fs/btrfs/volumes.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <linux/raid/pq.h>
  29 #include <linux/semaphore.h>
  30 #include <asm/div64.h>
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43
  44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  45                                 struct btrfs_root *root,
  46                                 struct btrfs_device *device);
  47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  49 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
  50 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  51
  52 static DEFINE_MUTEX(uuid_mutex);
  53 static LIST_HEAD(fs_uuids);
  54
  55 static void lock_chunks(struct btrfs_root *root)
  56 {
  57         mutex_lock(&root->fs_info->chunk_mutex);
  58 }
  59
  60 static void unlock_chunks(struct btrfs_root *root)
  61 {
  62         mutex_unlock(&root->fs_info->chunk_mutex);
  63 }
  64
  65 static struct btrfs_fs_devices *__alloc_fs_devices(void)
  66 {
  67         struct btrfs_fs_devices *fs_devs;
  68
  69         fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
  70         if (!fs_devs)
  71                 return ERR_PTR(-ENOMEM);
  72
  73         mutex_init(&fs_devs->device_list_mutex);
  74
  75         INIT_LIST_HEAD(&fs_devs->devices);
  76         INIT_LIST_HEAD(&fs_devs->alloc_list);
  77         INIT_LIST_HEAD(&fs_devs->list);
  78
  79         return fs_devs;
  80 }
  81
  82 /**
  83  * alloc_fs_devices - allocate struct btrfs_fs_devices
  84  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
  85  *              generated.
  86  *
  87  * Return: a pointer to a new &struct btrfs_fs_devices on success;
  88  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
  89  * can be destroyed with kfree() right away.
  90  */
  91 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
  92 {
  93         struct btrfs_fs_devices *fs_devs;
  94
  95         fs_devs = __alloc_fs_devices();
  96         if (IS_ERR(fs_devs))
  97                 return fs_devs;
  98
  99         if (fsid)
 100                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 101         else
 102                 generate_random_uuid(fs_devs->fsid);
 103
 104         return fs_devs;
 105 }
 106
 107 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 108 {
 109         struct btrfs_device *device;
 110         WARN_ON(fs_devices->opened);
 111         while (!list_empty(&fs_devices->devices)) {
 112                 device = list_entry(fs_devices->devices.next,
 113                                     struct btrfs_device, dev_list);
 114                 list_del(&device->dev_list);
 115                 rcu_string_free(device->name);
 116                 kfree(device);
 117         }
 118         kfree(fs_devices);
 119 }
 120
 121 static void btrfs_kobject_uevent(struct block_device *bdev,
 122                                  enum kobject_action action)
 123 {
 124         int ret;
 125
 126         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
 127         if (ret)
 128                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
 129                         action,
 130                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
 131                         &disk_to_dev(bdev->bd_disk)->kobj);
 132 }
 133
 134 void btrfs_cleanup_fs_uuids(void)
 135 {
 136         struct btrfs_fs_devices *fs_devices;
 137
 138         while (!list_empty(&fs_uuids)) {
 139                 fs_devices = list_entry(fs_uuids.next,
 140                                         struct btrfs_fs_devices, list);
 141                 list_del(&fs_devices->list);
 142                 free_fs_devices(fs_devices);
 143         }
 144 }
 145
 146 static struct btrfs_device *__alloc_device(void)
 147 {
 148         struct btrfs_device *dev;
 149
 150         dev = kzalloc(sizeof(*dev), GFP_NOFS);
 151         if (!dev)
 152                 return ERR_PTR(-ENOMEM);
 153
 154         INIT_LIST_HEAD(&dev->dev_list);
 155         INIT_LIST_HEAD(&dev->dev_alloc_list);
 156
 157         spin_lock_init(&dev->io_lock);
 158
 159         spin_lock_init(&dev->reada_lock);
 160         atomic_set(&dev->reada_in_flight, 0);
 161         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 162         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 163
 164         return dev;
 165 }
 166
 167 static noinline struct btrfs_device *__find_device(struct list_head *head,
 168                                                    u64 devid, u8 *uuid)
 169 {
 170         struct btrfs_device *dev;
 171
 172         list_for_each_entry(dev, head, dev_list) {
 173                 if (dev->devid == devid &&
 174                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 175                         return dev;
 176                 }
 177         }
 178         return NULL;
 179 }
 180
 181 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 182 {
 183         struct btrfs_fs_devices *fs_devices;
 184
 185         list_for_each_entry(fs_devices, &fs_uuids, list) {
 186                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 187                         return fs_devices;
 188         }
 189         return NULL;
 190 }
 191
 192 static int
 193 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 194                       int flush, struct block_device **bdev,
 195                       struct buffer_head **bh)
 196 {
 197         int ret;
 198
 199         *bdev = blkdev_get_by_path(device_path, flags, holder);
 200
 201         if (IS_ERR(*bdev)) {
 202                 ret = PTR_ERR(*bdev);
 203                 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
 204                 goto error;
 205         }
 206
 207         if (flush)
 208                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 209         ret = set_blocksize(*bdev, 4096);
 210         if (ret) {
 211                 blkdev_put(*bdev, flags);
 212                 goto error;
 213         }
 214         invalidate_bdev(*bdev);
 215         *bh = btrfs_read_dev_super(*bdev);
 216         if (!*bh) {
 217                 ret = -EINVAL;
 218                 blkdev_put(*bdev, flags);
 219                 goto error;
 220         }
 221
 222         return 0;
 223
 224 error:
 225         *bdev = NULL;
 226         *bh = NULL;
 227         return ret;
 228 }
 229
 230 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 231                         struct bio *head, struct bio *tail)
 232 {
 233
 234         struct bio *old_head;
 235
 236         old_head = pending_bios->head;
 237         pending_bios->head = head;
 238         if (pending_bios->tail)
 239                 tail->bi_next = old_head;
 240         else
 241                 pending_bios->tail = tail;
 242 }
 243
 244 /*
 245  * we try to collect pending bios for a device so we don't get a large
 246  * number of procs sending bios down to the same device.  This greatly
 247  * improves the schedulers ability to collect and merge the bios.
 248  *
 249  * But, it also turns into a long list of bios to process and that is sure
 250  * to eventually make the worker thread block.  The solution here is to
 251  * make some progress and then put this work struct back at the end of
 252  * the list if the block device is congested.  This way, multiple devices
 253  * can make progress from a single worker thread.
 254  */
 255 static noinline void run_scheduled_bios(struct btrfs_device *device)
 256 {
 257         struct bio *pending;
 258         struct backing_dev_info *bdi;
 259         struct btrfs_fs_info *fs_info;
 260         struct btrfs_pending_bios *pending_bios;
 261         struct bio *tail;
 262         struct bio *cur;
 263         int again = 0;
 264         unsigned long num_run;
 265         unsigned long batch_run = 0;
 266         unsigned long limit;
 267         unsigned long last_waited = 0;
 268         int force_reg = 0;
 269         int sync_pending = 0;
 270         struct blk_plug plug;
 271
 272         /*
 273          * this function runs all the bios we've collected for
 274          * a particular device.  We don't want to wander off to
 275          * another device without first sending all of these down.
 276          * So, setup a plug here and finish it off before we return
 277          */
 278         blk_start_plug(&plug);
 279
 280         bdi = blk_get_backing_dev_info(device->bdev);
 281         fs_info = device->dev_root->fs_info;
 282         limit = btrfs_async_submit_limit(fs_info);
 283         limit = limit * 2 / 3;
 284
 285 loop:
 286         spin_lock(&device->io_lock);
 287
 288 loop_lock:
 289         num_run = 0;
 290
 291         /* take all the bios off the list at once and process them
 292          * later on (without the lock held).  But, remember the
 293          * tail and other pointers so the bios can be properly reinserted
 294          * into the list if we hit congestion
 295          */
 296         if (!force_reg && device->pending_sync_bios.head) {
 297                 pending_bios = &device->pending_sync_bios;
 298                 force_reg = 1;
 299         } else {
 300                 pending_bios = &device->pending_bios;
 301                 force_reg = 0;
 302         }
 303
 304         pending = pending_bios->head;
 305         tail = pending_bios->tail;
 306         WARN_ON(pending && !tail);
 307
 308         /*
 309          * if pending was null this time around, no bios need processing
 310          * at all and we can stop.  Otherwise it'll loop back up again
 311          * and do an additional check so no bios are missed.
 312          *
 313          * device->running_pending is used to synchronize with the
 314          * schedule_bio code.
 315          */
 316         if (device->pending_sync_bios.head == NULL &&
 317             device->pending_bios.head == NULL) {
 318                 again = 0;
 319                 device->running_pending = 0;
 320         } else {
 321                 again = 1;
 322                 device->running_pending = 1;
 323         }
 324
 325         pending_bios->head = NULL;
 326         pending_bios->tail = NULL;
 327
 328         spin_unlock(&device->io_lock);
 329
 330         while (pending) {
 331
 332                 rmb();
 333                 /* we want to work on both lists, but do more bios on the
 334                  * sync list than the regular list
 335                  */
 336                 if ((num_run > 32 &&
 337                     pending_bios != &device->pending_sync_bios &&
 338                     device->pending_sync_bios.head) ||
 339                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 340                     device->pending_bios.head)) {
 341                         spin_lock(&device->io_lock);
 342                         requeue_list(pending_bios, pending, tail);
 343                         goto loop_lock;
 344                 }
 345
 346                 cur = pending;
 347                 pending = pending->bi_next;
 348                 cur->bi_next = NULL;
 349
 350                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 351                     waitqueue_active(&fs_info->async_submit_wait))
 352                         wake_up(&fs_info->async_submit_wait);
 353
 354                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 355
 356                 /*
 357                  * if we're doing the sync list, record that our
 358                  * plug has some sync requests on it
 359                  *
 360                  * If we're doing the regular list and there are
 361                  * sync requests sitting around, unplug before
 362                  * we add more
 363                  */
 364                 if (pending_bios == &device->pending_sync_bios) {
 365                         sync_pending = 1;
 366                 } else if (sync_pending) {
 367                         blk_finish_plug(&plug);
 368                         blk_start_plug(&plug);
 369                         sync_pending = 0;
 370                 }
 371
 372                 btrfsic_submit_bio(cur->bi_rw, cur);
 373                 num_run++;
 374                 batch_run++;
 375                 if (need_resched())
 376                         cond_resched();
 377
 378                 /*
 379                  * we made progress, there is more work to do and the bdi
 380                  * is now congested.  Back off and let other work structs
 381                  * run instead
 382                  */
 383                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 384                     fs_info->fs_devices->open_devices > 1) {
 385                         struct io_context *ioc;
 386
 387                         ioc = current->io_context;
 388
 389                         /*
 390                          * the main goal here is that we don't want to
 391                          * block if we're going to be able to submit
 392                          * more requests without blocking.
 393                          *
 394                          * This code does two great things, it pokes into
 395                          * the elevator code from a filesystem _and_
 396                          * it makes assumptions about how batching works.
 397                          */
 398                         if (ioc && ioc->nr_batch_requests > 0 &&
 399                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 400                             (last_waited == 0 ||
 401                              ioc->last_waited == last_waited)) {
 402                                 /*
 403                                  * we want to go through our batch of
 404                                  * requests and stop.  So, we copy out
 405                                  * the ioc->last_waited time and test
 406                                  * against it before looping
 407                                  */
 408                                 last_waited = ioc->last_waited;
 409                                 if (need_resched())
 410                                         cond_resched();
 411                                 continue;
 412                         }
 413                         spin_lock(&device->io_lock);
 414                         requeue_list(pending_bios, pending, tail);
 415                         device->running_pending = 1;
 416
 417                         spin_unlock(&device->io_lock);
 418                         btrfs_queue_work(fs_info->submit_workers,
 419                                          &device->work);
 420                         goto done;
 421                 }
 422                 /* unplug every 64 requests just for good measure */
 423                 if (batch_run % 64 == 0) {
 424                         blk_finish_plug(&plug);
 425                         blk_start_plug(&plug);
 426                         sync_pending = 0;
 427                 }
 428         }
 429
 430         cond_resched();
 431         if (again)
 432                 goto loop;
 433
 434         spin_lock(&device->io_lock);
 435         if (device->pending_bios.head || device->pending_sync_bios.head)
 436                 goto loop_lock;
 437         spin_unlock(&device->io_lock);
 438
 439 done:
 440         blk_finish_plug(&plug);
 441 }
 442
 443 static void pending_bios_fn(struct btrfs_work *work)
 444 {
 445         struct btrfs_device *device;
 446
 447         device = container_of(work, struct btrfs_device, work);
 448         run_scheduled_bios(device);
 449 }
 450
 451 /*
 452  * Add new device to list of registered devices
 453  *
 454  * Returns:
 455  * 1   - first time device is seen
 456  * 0   - device already known
 457  * < 0 - error
 458  */
 459 static noinline int device_list_add(const char *path,
 460                            struct btrfs_super_block *disk_super,
 461                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 462 {
 463         struct btrfs_device *device;
 464         struct btrfs_fs_devices *fs_devices;
 465         struct rcu_string *name;
 466         int ret = 0;
 467         u64 found_transid = btrfs_super_generation(disk_super);
 468
 469         fs_devices = find_fsid(disk_super->fsid);
 470         if (!fs_devices) {
 471                 fs_devices = alloc_fs_devices(disk_super->fsid);
 472                 if (IS_ERR(fs_devices))
 473                         return PTR_ERR(fs_devices);
 474
 475                 list_add(&fs_devices->list, &fs_uuids);
 476                 fs_devices->latest_devid = devid;
 477                 fs_devices->latest_trans = found_transid;
 478
 479                 device = NULL;
 480         } else {
 481                 device = __find_device(&fs_devices->devices, devid,
 482                                        disk_super->dev_item.uuid);
 483         }
 484         if (!device) {
 485                 if (fs_devices->opened)
 486                         return -EBUSY;
 487
 488                 device = btrfs_alloc_device(NULL, &devid,
 489                                             disk_super->dev_item.uuid);
 490                 if (IS_ERR(device)) {
 491                         /* we can safely leave the fs_devices entry around */
 492                         return PTR_ERR(device);
 493                 }
 494
 495                 name = rcu_string_strdup(path, GFP_NOFS);
 496                 if (!name) {
 497                         kfree(device);
 498                         return -ENOMEM;
 499                 }
 500                 rcu_assign_pointer(device->name, name);
 501
 502                 mutex_lock(&fs_devices->device_list_mutex);
 503                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 504                 fs_devices->num_devices++;
 505                 mutex_unlock(&fs_devices->device_list_mutex);
 506
 507                 ret = 1;
 508                 device->fs_devices = fs_devices;
 509         } else if (!device->name || strcmp(device->name->str, path)) {
 510                 name = rcu_string_strdup(path, GFP_NOFS);
 511                 if (!name)
 512                         return -ENOMEM;
 513                 rcu_string_free(device->name);
 514                 rcu_assign_pointer(device->name, name);
 515                 if (device->missing) {
 516                         fs_devices->missing_devices--;
 517                         device->missing = 0;
 518                 }
 519         }
 520
 521         if (found_transid > fs_devices->latest_trans) {
 522                 fs_devices->latest_devid = devid;
 523                 fs_devices->latest_trans = found_transid;
 524         }
 525         *fs_devices_ret = fs_devices;
 526
 527         return ret;
 528 }
 529
 530 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 531 {
 532         struct btrfs_fs_devices *fs_devices;
 533         struct btrfs_device *device;
 534         struct btrfs_device *orig_dev;
 535
 536         fs_devices = alloc_fs_devices(orig->fsid);
 537         if (IS_ERR(fs_devices))
 538                 return fs_devices;
 539
 540         fs_devices->latest_devid = orig->latest_devid;
 541         fs_devices->latest_trans = orig->latest_trans;
 542         fs_devices->total_devices = orig->total_devices;
 543
 544         /* We have held the volume lock, it is safe to get the devices. */
 545         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 546                 struct rcu_string *name;
 547
 548                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
 549                                             orig_dev->uuid);
 550                 if (IS_ERR(device))
 551                         goto error;
 552
 553                 /*
 554                  * This is ok to do without rcu read locked because we hold the
 555                  * uuid mutex so nothing we touch in here is going to disappear.
 556                  */
 557                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 558                 if (!name) {
 559                         kfree(device);
 560                         goto error;
 561                 }
 562                 rcu_assign_pointer(device->name, name);
 563
 564                 list_add(&device->dev_list, &fs_devices->devices);
 565                 device->fs_devices = fs_devices;
 566                 fs_devices->num_devices++;
 567         }
 568         return fs_devices;
 569 error:
 570         free_fs_devices(fs_devices);
 571         return ERR_PTR(-ENOMEM);
 572 }
 573
 574 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 575                                struct btrfs_fs_devices *fs_devices, int step)
 576 {
 577         struct btrfs_device *device, *next;
 578
 579         struct block_device *latest_bdev = NULL;
 580         u64 latest_devid = 0;
 581         u64 latest_transid = 0;
 582
 583         mutex_lock(&uuid_mutex);
 584 again:
 585         /* This is the initialized path, it is safe to release the devices. */
 586         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 587                 if (device->in_fs_metadata) {
 588                         if (!device->is_tgtdev_for_dev_replace &&
 589                             (!latest_transid ||
 590                              device->generation > latest_transid)) {
 591                                 latest_devid = device->devid;
 592                                 latest_transid = device->generation;
 593                                 latest_bdev = device->bdev;
 594                         }
 595                         continue;
 596                 }
 597
 598                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 599                         /*
 600                          * In the first step, keep the device which has
 601                          * the correct fsid and the devid that is used
 602                          * for the dev_replace procedure.
 603                          * In the second step, the dev_replace state is
 604                          * read from the device tree and it is known
 605                          * whether the procedure is really active or
 606                          * not, which means whether this device is
 607                          * used or whether it should be removed.
 608                          */
 609                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 610                                 continue;
 611                         }
 612                 }
 613                 if (device->bdev) {
 614                         blkdev_put(device->bdev, device->mode);
 615                         device->bdev = NULL;
 616                         fs_devices->open_devices--;
 617                 }
 618                 if (device->writeable) {
 619                         list_del_init(&device->dev_alloc_list);
 620                         device->writeable = 0;
 621                         if (!device->is_tgtdev_for_dev_replace)
 622                                 fs_devices->rw_devices--;
 623                 }
 624                 list_del_init(&device->dev_list);
 625                 fs_devices->num_devices--;
 626                 rcu_string_free(device->name);
 627                 kfree(device);
 628         }
 629
 630         if (fs_devices->seed) {
 631                 fs_devices = fs_devices->seed;
 632                 goto again;
 633         }
 634
 635         fs_devices->latest_bdev = latest_bdev;
 636         fs_devices->latest_devid = latest_devid;
 637         fs_devices->latest_trans = latest_transid;
 638
 639         mutex_unlock(&uuid_mutex);
 640 }
 641
 642 static void __free_device(struct work_struct *work)
 643 {
 644         struct btrfs_device *device;
 645
 646         device = container_of(work, struct btrfs_device, rcu_work);
 647
 648         if (device->bdev)
 649                 blkdev_put(device->bdev, device->mode);
 650
 651         rcu_string_free(device->name);
 652         kfree(device);
 653 }
 654
 655 static void free_device(struct rcu_head *head)
 656 {
 657         struct btrfs_device *device;
 658
 659         device = container_of(head, struct btrfs_device, rcu);
 660
 661         INIT_WORK(&device->rcu_work, __free_device);
 662         schedule_work(&device->rcu_work);
 663 }
 664
 665 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 666 {
 667         struct btrfs_device *device;
 668
 669         if (--fs_devices->opened > 0)
 670                 return 0;
 671
 672         mutex_lock(&fs_devices->device_list_mutex);
 673         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 674                 struct btrfs_device *new_device;
 675                 struct rcu_string *name;
 676
 677                 if (device->bdev)
 678                         fs_devices->open_devices--;
 679
 680                 if (device->writeable &&
 681                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 682                         list_del_init(&device->dev_alloc_list);
 683                         fs_devices->rw_devices--;
 684                 }
 685
 686                 if (device->can_discard)
 687                         fs_devices->num_can_discard--;
 688                 if (device->missing)
 689                         fs_devices->missing_devices--;
 690
 691                 new_device = btrfs_alloc_device(NULL, &device->devid,
 692                                                 device->uuid);
 693                 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
 694
 695                 /* Safe because we are under uuid_mutex */
 696                 if (device->name) {
 697                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 698                         BUG_ON(!name); /* -ENOMEM */
 699                         rcu_assign_pointer(new_device->name, name);
 700                 }
 701
 702                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 703                 new_device->fs_devices = device->fs_devices;
 704
 705                 call_rcu(&device->rcu, free_device);
 706         }
 707         mutex_unlock(&fs_devices->device_list_mutex);
 708
 709         WARN_ON(fs_devices->open_devices);
 710         WARN_ON(fs_devices->rw_devices);
 711         fs_devices->opened = 0;
 712         fs_devices->seeding = 0;
 713
 714         return 0;
 715 }
 716
 717 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 718 {
 719         struct btrfs_fs_devices *seed_devices = NULL;
 720         int ret;
 721
 722         mutex_lock(&uuid_mutex);
 723         ret = __btrfs_close_devices(fs_devices);
 724         if (!fs_devices->opened) {
 725                 seed_devices = fs_devices->seed;
 726                 fs_devices->seed = NULL;
 727         }
 728         mutex_unlock(&uuid_mutex);
 729
 730         while (seed_devices) {
 731                 fs_devices = seed_devices;
 732                 seed_devices = fs_devices->seed;
 733                 __btrfs_close_devices(fs_devices);
 734                 free_fs_devices(fs_devices);
 735         }
 736         /*
 737          * Wait for rcu kworkers under __btrfs_close_devices
 738          * to finish all blkdev_puts so device is really
 739          * free when umount is done.
 740          */
 741         rcu_barrier();
 742         return ret;
 743 }
 744
 745 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 746                                 fmode_t flags, void *holder)
 747 {
 748         struct request_queue *q;
 749         struct block_device *bdev;
 750         struct list_head *head = &fs_devices->devices;
 751         struct btrfs_device *device;
 752         struct block_device *latest_bdev = NULL;
 753         struct buffer_head *bh;
 754         struct btrfs_super_block *disk_super;
 755         u64 latest_devid = 0;
 756         u64 latest_transid = 0;
 757         u64 devid;
 758         int seeding = 1;
 759         int ret = 0;
 760
 761         flags |= FMODE_EXCL;
 762
 763         list_for_each_entry(device, head, dev_list) {
 764                 if (device->bdev)
 765                         continue;
 766                 if (!device->name)
 767                         continue;
 768
 769                 /* Just open everything we can; ignore failures here */
 770                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 771                                             &bdev, &bh))
 772                         continue;
 773
 774                 disk_super = (struct btrfs_super_block *)bh->b_data;
 775                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 776                 if (devid != device->devid)
 777                         goto error_brelse;
 778
 779                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 780                            BTRFS_UUID_SIZE))
 781                         goto error_brelse;
 782
 783                 device->generation = btrfs_super_generation(disk_super);
 784                 if (!latest_transid || device->generation > latest_transid) {
 785                         latest_devid = devid;
 786                         latest_transid = device->generation;
 787                         latest_bdev = bdev;
 788                 }
 789
 790                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 791                         device->writeable = 0;
 792                 } else {
 793                         device->writeable = !bdev_read_only(bdev);
 794                         seeding = 0;
 795                 }
 796
 797                 q = bdev_get_queue(bdev);
 798                 if (blk_queue_discard(q)) {
 799                         device->can_discard = 1;
 800                         fs_devices->num_can_discard++;
 801                 }
 802
 803                 device->bdev = bdev;
 804                 device->in_fs_metadata = 0;
 805                 device->mode = flags;
 806
 807                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 808                         fs_devices->rotating = 1;
 809
 810                 fs_devices->open_devices++;
 811                 if (device->writeable &&
 812                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
 813                         fs_devices->rw_devices++;
 814                         list_add(&device->dev_alloc_list,
 815                                  &fs_devices->alloc_list);
 816                 }
 817                 brelse(bh);
 818                 continue;
 819
 820 error_brelse:
 821                 brelse(bh);
 822                 blkdev_put(bdev, flags);
 823                 continue;
 824         }
 825         if (fs_devices->open_devices == 0) {
 826                 ret = -EINVAL;
 827                 goto out;
 828         }
 829         fs_devices->seeding = seeding;
 830         fs_devices->opened = 1;
 831         fs_devices->latest_bdev = latest_bdev;
 832         fs_devices->latest_devid = latest_devid;
 833         fs_devices->latest_trans = latest_transid;
 834         fs_devices->total_rw_bytes = 0;
 835 out:
 836         return ret;
 837 }
 838
 839 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 840                        fmode_t flags, void *holder)
 841 {
 842         int ret;
 843
 844         mutex_lock(&uuid_mutex);
 845         if (fs_devices->opened) {
 846                 fs_devices->opened++;
 847                 ret = 0;
 848         } else {
 849                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 850         }
 851         mutex_unlock(&uuid_mutex);
 852         return ret;
 853 }
 854
 855 /*
 856  * Look for a btrfs signature on a device. This may be called out of the mount path
 857  * and we are not allowed to call set_blocksize during the scan. The superblock
 858  * is read via pagecache
 859  */
 860 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 861                           struct btrfs_fs_devices **fs_devices_ret)
 862 {
 863         struct btrfs_super_block *disk_super;
 864         struct block_device *bdev;
 865         struct page *page;
 866         void *p;
 867         int ret = -EINVAL;
 868         u64 devid;
 869         u64 transid;
 870         u64 total_devices;
 871         u64 bytenr;
 872         pgoff_t index;
 873
 874         /*
 875          * we would like to check all the supers, but that would make
 876          * a btrfs mount succeed after a mkfs from a different FS.
 877          * So, we need to add a special mount option to scan for
 878          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 879          */
 880         bytenr = btrfs_sb_offset(0);
 881         flags |= FMODE_EXCL;
 882         mutex_lock(&uuid_mutex);
 883
 884         bdev = blkdev_get_by_path(path, flags, holder);
 885
 886         if (IS_ERR(bdev)) {
 887                 ret = PTR_ERR(bdev);
 888                 goto error;
 889         }
 890
 891         /* make sure our super fits in the device */
 892         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 893                 goto error_bdev_put;
 894
 895         /* make sure our super fits in the page */
 896         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 897                 goto error_bdev_put;
 898
 899         /* make sure our super doesn't straddle pages on disk */
 900         index = bytenr >> PAGE_CACHE_SHIFT;
 901         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 902                 goto error_bdev_put;
 903
 904         /* pull in the page with our super */
 905         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 906                                    index, GFP_NOFS);
 907
 908         if (IS_ERR_OR_NULL(page))
 909                 goto error_bdev_put;
 910
 911         p = kmap(page);
 912
 913         /* align our pointer to the offset of the super block */
 914         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 915
 916         if (btrfs_super_bytenr(disk_super) != bytenr ||
 917             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
 918                 goto error_unmap;
 919
 920         devid = btrfs_stack_device_id(&disk_super->dev_item);
 921         transid = btrfs_super_generation(disk_super);
 922         total_devices = btrfs_super_num_devices(disk_super);
 923
 924         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 925         if (ret > 0) {
 926                 if (disk_super->label[0]) {
 927                         if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 928                                 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 929                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
 930                 } else {
 931                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
 932                 }
 933
 934                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
 935                 ret = 0;
 936         }
 937         if (!ret && fs_devices_ret)
 938                 (*fs_devices_ret)->total_devices = total_devices;
 939
 940 error_unmap:
 941         kunmap(page);
 942         page_cache_release(page);
 943
 944 error_bdev_put:
 945         blkdev_put(bdev, flags);
 946 error:
 947         mutex_unlock(&uuid_mutex);
 948         return ret;
 949 }
 950
 951 /* helper to account the used device space in the range */
 952 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 953                                    u64 end, u64 *length)
 954 {
 955         struct btrfs_key key;
 956         struct btrfs_root *root = device->dev_root;
 957         struct btrfs_dev_extent *dev_extent;
 958         struct btrfs_path *path;
 959         u64 extent_end;
 960         int ret;
 961         int slot;
 962         struct extent_buffer *l;
 963
 964         *length = 0;
 965
 966         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 967                 return 0;
 968
 969         path = btrfs_alloc_path();
 970         if (!path)
 971                 return -ENOMEM;
 972         path->reada = 2;
 973
 974         key.objectid = device->devid;
 975         key.offset = start;
 976         key.type = BTRFS_DEV_EXTENT_KEY;
 977
 978         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 979         if (ret < 0)
 980                 goto out;
 981         if (ret > 0) {
 982                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 983                 if (ret < 0)
 984                         goto out;
 985         }
 986
 987         while (1) {
 988                 l = path->nodes[0];
 989                 slot = path->slots[0];
 990                 if (slot >= btrfs_header_nritems(l)) {
 991                         ret = btrfs_next_leaf(root, path);
 992                         if (ret == 0)
 993                                 continue;
 994                         if (ret < 0)
 995                                 goto out;
 996
 997                         break;
 998                 }
 999                 btrfs_item_key_to_cpu(l, &key, slot);
1000
1001                 if (key.objectid < device->devid)
1002                         goto next;
1003
1004                 if (key.objectid > device->devid)
1005                         break;
1006
1007                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1008                         goto next;
1009
1010                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1011                 extent_end = key.offset + btrfs_dev_extent_length(l,
1012                                                                   dev_extent);
1013                 if (key.offset <= start && extent_end > end) {
1014                         *length = end - start + 1;
1015                         break;
1016                 } else if (key.offset <= start && extent_end > start)
1017                         *length += extent_end - start;
1018                 else if (key.offset > start && extent_end <= end)
1019                         *length += extent_end - key.offset;
1020                 else if (key.offset > start && key.offset <= end) {
1021                         *length += end - key.offset + 1;
1022                         break;
1023                 } else if (key.offset > end)
1024                         break;
1025
1026 next:
1027                 path->slots[0]++;
1028         }
1029         ret = 0;
1030 out:
1031         btrfs_free_path(path);
1032         return ret;
1033 }
1034
1035 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1036                                    struct btrfs_device *device,
1037                                    u64 *start, u64 len)
1038 {
1039         struct extent_map *em;
1040         int ret = 0;
1041
1042         list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1043                 struct map_lookup *map;
1044                 int i;
1045
1046                 map = (struct map_lookup *)em->bdev;
1047                 for (i = 0; i < map->num_stripes; i++) {
1048                         if (map->stripes[i].dev != device)
1049                                 continue;
1050                         if (map->stripes[i].physical >= *start + len ||
1051                             map->stripes[i].physical + em->orig_block_len <=
1052                             *start)
1053                                 continue;
1054                         *start = map->stripes[i].physical +
1055                                 em->orig_block_len;
1056                         ret = 1;
1057                 }
1058         }
1059
1060         return ret;
1061 }
1062
1063
1064 /*
1065  * find_free_dev_extent - find free space in the specified device
1066  * @device:     the device which we search the free space in
1067  * @num_bytes:  the size of the free space that we need
1068  * @start:      store the start of the free space.
1069  * @len:        the size of the free space. that we find, or the size of the max
1070  *              free space if we don't find suitable free space
1071  *
1072  * this uses a pretty simple search, the expectation is that it is
1073  * called very infrequently and that a given device has a small number
1074  * of extents
1075  *
1076  * @start is used to store the start of the free space if we find. But if we
1077  * don't find suitable free space, it will be used to store the start position
1078  * of the max free space.
1079  *
1080  * @len is used to store the size of the free space that we find.
1081  * But if we don't find suitable free space, it is used to store the size of
1082  * the max free space.
1083  */
1084 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1085                          struct btrfs_device *device, u64 num_bytes,
1086                          u64 *start, u64 *len)
1087 {
1088         struct btrfs_key key;
1089         struct btrfs_root *root = device->dev_root;
1090         struct btrfs_dev_extent *dev_extent;
1091         struct btrfs_path *path;
1092         u64 hole_size;
1093         u64 max_hole_start;
1094         u64 max_hole_size;
1095         u64 extent_end;
1096         u64 search_start;
1097         u64 search_end = device->total_bytes;
1098         int ret;
1099         int slot;
1100         struct extent_buffer *l;
1101
1102         /* FIXME use last free of some kind */
1103
1104         /* we don't want to overwrite the superblock on the drive,
1105          * so we make sure to start at an offset of at least 1MB
1106          */
1107         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1108
1109         path = btrfs_alloc_path();
1110         if (!path)
1111                 return -ENOMEM;
1112 again:
1113         max_hole_start = search_start;
1114         max_hole_size = 0;
1115         hole_size = 0;
1116
1117         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1118                 ret = -ENOSPC;
1119                 goto out;
1120         }
1121
1122         path->reada = 2;
1123         path->search_commit_root = 1;
1124         path->skip_locking = 1;
1125
1126         key.objectid = device->devid;
1127         key.offset = search_start;
1128         key.type = BTRFS_DEV_EXTENT_KEY;
1129
1130         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1131         if (ret < 0)
1132                 goto out;
1133         if (ret > 0) {
1134                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1135                 if (ret < 0)
1136                         goto out;
1137         }
1138
1139         while (1) {
1140                 l = path->nodes[0];
1141                 slot = path->slots[0];
1142                 if (slot >= btrfs_header_nritems(l)) {
1143                         ret = btrfs_next_leaf(root, path);
1144                         if (ret == 0)
1145                                 continue;
1146                         if (ret < 0)
1147                                 goto out;
1148
1149                         break;
1150                 }
1151                 btrfs_item_key_to_cpu(l, &key, slot);
1152
1153                 if (key.objectid < device->devid)
1154                         goto next;
1155
1156                 if (key.objectid > device->devid)
1157                         break;
1158
1159                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1160                         goto next;
1161
1162                 if (key.offset > search_start) {
1163                         hole_size = key.offset - search_start;
1164
1165                         /*
1166                          * Have to check before we set max_hole_start, otherwise
1167                          * we could end up sending back this offset anyway.
1168                          */
1169                         if (contains_pending_extent(trans, device,
1170                                                     &search_start,
1171                                                     hole_size))
1172                                 hole_size = 0;
1173
1174                         if (hole_size > max_hole_size) {
1175                                 max_hole_start = search_start;
1176                                 max_hole_size = hole_size;
1177                         }
1178
1179                         /*
1180                          * If this free space is greater than which we need,
1181                          * it must be the max free space that we have found
1182                          * until now, so max_hole_start must point to the start
1183                          * of this free space and the length of this free space
1184                          * is stored in max_hole_size. Thus, we return
1185                          * max_hole_start and max_hole_size and go back to the
1186                          * caller.
1187                          */
1188                         if (hole_size >= num_bytes) {
1189                                 ret = 0;
1190                                 goto out;
1191                         }
1192                 }
1193
1194                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1195                 extent_end = key.offset + btrfs_dev_extent_length(l,
1196                                                                   dev_extent);
1197                 if (extent_end > search_start)
1198                         search_start = extent_end;
1199 next:
1200                 path->slots[0]++;
1201                 cond_resched();
1202         }
1203
1204         /*
1205          * At this point, search_start should be the end of
1206          * allocated dev extents, and when shrinking the device,
1207          * search_end may be smaller than search_start.
1208          */
1209         if (search_end > search_start)
1210                 hole_size = search_end - search_start;
1211
1212         if (hole_size > max_hole_size) {
1213                 max_hole_start = search_start;
1214                 max_hole_size = hole_size;
1215         }
1216
1217         if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1218                 btrfs_release_path(path);
1219                 goto again;
1220         }
1221
1222         /* See above. */
1223         if (hole_size < num_bytes)
1224                 ret = -ENOSPC;
1225         else
1226                 ret = 0;
1227
1228 out:
1229         btrfs_free_path(path);
1230         *start = max_hole_start;
1231         if (len)
1232                 *len = max_hole_size;
1233         return ret;
1234 }
1235
1236 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1237                           struct btrfs_device *device,
1238                           u64 start)
1239 {
1240         int ret;
1241         struct btrfs_path *path;
1242         struct btrfs_root *root = device->dev_root;
1243         struct btrfs_key key;
1244         struct btrfs_key found_key;
1245         struct extent_buffer *leaf = NULL;
1246         struct btrfs_dev_extent *extent = NULL;
1247
1248         path = btrfs_alloc_path();
1249         if (!path)
1250                 return -ENOMEM;
1251
1252         key.objectid = device->devid;
1253         key.offset = start;
1254         key.type = BTRFS_DEV_EXTENT_KEY;
1255 again:
1256         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1257         if (ret > 0) {
1258                 ret = btrfs_previous_item(root, path, key.objectid,
1259                                           BTRFS_DEV_EXTENT_KEY);
1260                 if (ret)
1261                         goto out;
1262                 leaf = path->nodes[0];
1263                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1264                 extent = btrfs_item_ptr(leaf, path->slots[0],
1265                                         struct btrfs_dev_extent);
1266                 BUG_ON(found_key.offset > start || found_key.offset +
1267                        btrfs_dev_extent_length(leaf, extent) < start);
1268                 key = found_key;
1269                 btrfs_release_path(path);
1270                 goto again;
1271         } else if (ret == 0) {
1272                 leaf = path->nodes[0];
1273                 extent = btrfs_item_ptr(leaf, path->slots[0],
1274                                         struct btrfs_dev_extent);
1275         } else {
1276                 btrfs_error(root->fs_info, ret, "Slot search failed");
1277                 goto out;
1278         }
1279
1280         if (device->bytes_used > 0) {
1281                 u64 len = btrfs_dev_extent_length(leaf, extent);
1282                 device->bytes_used -= len;
1283                 spin_lock(&root->fs_info->free_chunk_lock);
1284                 root->fs_info->free_chunk_space += len;
1285                 spin_unlock(&root->fs_info->free_chunk_lock);
1286         }
1287         ret = btrfs_del_item(trans, root, path);
1288         if (ret) {
1289                 btrfs_error(root->fs_info, ret,
1290                             "Failed to remove dev extent item");
1291         }
1292 out:
1293         btrfs_free_path(path);
1294         return ret;
1295 }
1296
1297 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1298                                   struct btrfs_device *device,
1299                                   u64 chunk_tree, u64 chunk_objectid,
1300                                   u64 chunk_offset, u64 start, u64 num_bytes)
1301 {
1302         int ret;
1303         struct btrfs_path *path;
1304         struct btrfs_root *root = device->dev_root;
1305         struct btrfs_dev_extent *extent;
1306         struct extent_buffer *leaf;
1307         struct btrfs_key key;
1308
1309         WARN_ON(!device->in_fs_metadata);
1310         WARN_ON(device->is_tgtdev_for_dev_replace);
1311         path = btrfs_alloc_path();
1312         if (!path)
1313                 return -ENOMEM;
1314
1315         key.objectid = device->devid;
1316         key.offset = start;
1317         key.type = BTRFS_DEV_EXTENT_KEY;
1318         ret = btrfs_insert_empty_item(trans, root, path, &key,
1319                                       sizeof(*extent));
1320         if (ret)
1321                 goto out;
1322
1323         leaf = path->nodes[0];
1324         extent = btrfs_item_ptr(leaf, path->slots[0],
1325                                 struct btrfs_dev_extent);
1326         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1327         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1328         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1329
1330         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1331                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1332
1333         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1334         btrfs_mark_buffer_dirty(leaf);
1335 out:
1336         btrfs_free_path(path);
1337         return ret;
1338 }
1339
1340 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1341 {
1342         struct extent_map_tree *em_tree;
1343         struct extent_map *em;
1344         struct rb_node *n;
1345         u64 ret = 0;
1346
1347         em_tree = &fs_info->mapping_tree.map_tree;
1348         read_lock(&em_tree->lock);
1349         n = rb_last(&em_tree->map);
1350         if (n) {
1351                 em = rb_entry(n, struct extent_map, rb_node);
1352                 ret = em->start + em->len;
1353         }
1354         read_unlock(&em_tree->lock);
1355
1356         return ret;
1357 }
1358
1359 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1360                                     u64 *devid_ret)
1361 {
1362         int ret;
1363         struct btrfs_key key;
1364         struct btrfs_key found_key;
1365         struct btrfs_path *path;
1366
1367         path = btrfs_alloc_path();
1368         if (!path)
1369                 return -ENOMEM;
1370
1371         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1372         key.type = BTRFS_DEV_ITEM_KEY;
1373         key.offset = (u64)-1;
1374
1375         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1376         if (ret < 0)
1377                 goto error;
1378
1379         BUG_ON(ret == 0); /* Corruption */
1380
1381         ret = btrfs_previous_item(fs_info->chunk_root, path,
1382                                   BTRFS_DEV_ITEMS_OBJECTID,
1383                                   BTRFS_DEV_ITEM_KEY);
1384         if (ret) {
1385                 *devid_ret = 1;
1386         } else {
1387                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1388                                       path->slots[0]);
1389                 *devid_ret = found_key.offset + 1;
1390         }
1391         ret = 0;
1392 error:
1393         btrfs_free_path(path);
1394         return ret;
1395 }
1396
1397 /*
1398  * the device information is stored in the chunk root
1399  * the btrfs_device struct should be fully filled in
1400  */
1401 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1402                             struct btrfs_root *root,
1403                             struct btrfs_device *device)
1404 {
1405         int ret;
1406         struct btrfs_path *path;
1407         struct btrfs_dev_item *dev_item;
1408         struct extent_buffer *leaf;
1409         struct btrfs_key key;
1410         unsigned long ptr;
1411
1412         root = root->fs_info->chunk_root;
1413
1414         path = btrfs_alloc_path();
1415         if (!path)
1416                 return -ENOMEM;
1417
1418         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1419         key.type = BTRFS_DEV_ITEM_KEY;
1420         key.offset = device->devid;
1421
1422         ret = btrfs_insert_empty_item(trans, root, path, &key,
1423                                       sizeof(*dev_item));
1424         if (ret)
1425                 goto out;
1426
1427         leaf = path->nodes[0];
1428         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1429
1430         btrfs_set_device_id(leaf, dev_item, device->devid);
1431         btrfs_set_device_generation(leaf, dev_item, 0);
1432         btrfs_set_device_type(leaf, dev_item, device->type);
1433         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1434         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1435         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1436         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1437         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1438         btrfs_set_device_group(leaf, dev_item, 0);
1439         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1440         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1441         btrfs_set_device_start_offset(leaf, dev_item, 0);
1442
1443         ptr = btrfs_device_uuid(dev_item);
1444         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1445         ptr = btrfs_device_fsid(dev_item);
1446         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1447         btrfs_mark_buffer_dirty(leaf);
1448
1449         ret = 0;
1450 out:
1451         btrfs_free_path(path);
1452         return ret;
1453 }
1454
1455 /*
1456  * Function to update ctime/mtime for a given device path.
1457  * Mainly used for ctime/mtime based probe like libblkid.
1458  */
1459 static void update_dev_time(char *path_name)
1460 {
1461         struct file *filp;
1462
1463         filp = filp_open(path_name, O_RDWR, 0);
1464         if (!filp)
1465                 return;
1466         file_update_time(filp);
1467         filp_close(filp, NULL);
1468         return;
1469 }
1470
1471 static int btrfs_rm_dev_item(struct btrfs_root *root,
1472                              struct btrfs_device *device)
1473 {
1474         int ret;
1475         struct btrfs_path *path;
1476         struct btrfs_key key;
1477         struct btrfs_trans_handle *trans;
1478
1479         root = root->fs_info->chunk_root;
1480
1481         path = btrfs_alloc_path();
1482         if (!path)
1483                 return -ENOMEM;
1484
1485         trans = btrfs_start_transaction(root, 0);
1486         if (IS_ERR(trans)) {
1487                 btrfs_free_path(path);
1488                 return PTR_ERR(trans);
1489         }
1490         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1491         key.type = BTRFS_DEV_ITEM_KEY;
1492         key.offset = device->devid;
1493         lock_chunks(root);
1494
1495         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1496         if (ret < 0)
1497                 goto out;
1498
1499         if (ret > 0) {
1500                 ret = -ENOENT;
1501                 goto out;
1502         }
1503
1504         ret = btrfs_del_item(trans, root, path);
1505         if (ret)
1506                 goto out;
1507 out:
1508         btrfs_free_path(path);
1509         unlock_chunks(root);
1510         btrfs_commit_transaction(trans, root);
1511         return ret;
1512 }
1513
1514 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1515 {
1516         struct btrfs_device *device;
1517         struct btrfs_device *next_device;
1518         struct block_device *bdev;
1519         struct buffer_head *bh = NULL;
1520         struct btrfs_super_block *disk_super;
1521         struct btrfs_fs_devices *cur_devices;
1522         u64 all_avail;
1523         u64 devid;
1524         u64 num_devices;
1525         u8 *dev_uuid;
1526         unsigned seq;
1527         int ret = 0;
1528         bool clear_super = false;
1529
1530         mutex_lock(&uuid_mutex);
1531
1532         do {
1533                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1534
1535                 all_avail = root->fs_info->avail_data_alloc_bits |
1536                             root->fs_info->avail_system_alloc_bits |
1537                             root->fs_info->avail_metadata_alloc_bits;
1538         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1539
1540         num_devices = root->fs_info->fs_devices->num_devices;
1541         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1542         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1543                 WARN_ON(num_devices < 1);
1544                 num_devices--;
1545         }
1546         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1547
1548         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1549                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1550                 goto out;
1551         }
1552
1553         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1554                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1555                 goto out;
1556         }
1557
1558         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1559             root->fs_info->fs_devices->rw_devices <= 2) {
1560                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1561                 goto out;
1562         }
1563         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1564             root->fs_info->fs_devices->rw_devices <= 3) {
1565                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1566                 goto out;
1567         }
1568
1569         if (strcmp(device_path, "missing") == 0) {
1570                 struct list_head *devices;
1571                 struct btrfs_device *tmp;
1572
1573                 device = NULL;
1574                 devices = &root->fs_info->fs_devices->devices;
1575                 /*
1576                  * It is safe to read the devices since the volume_mutex
1577                  * is held.
1578                  */
1579                 list_for_each_entry(tmp, devices, dev_list) {
1580                         if (tmp->in_fs_metadata &&
1581                             !tmp->is_tgtdev_for_dev_replace &&
1582                             !tmp->bdev) {
1583                                 device = tmp;
1584                                 break;
1585                         }
1586                 }
1587                 bdev = NULL;
1588                 bh = NULL;
1589                 disk_super = NULL;
1590                 if (!device) {
1591                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1592                         goto out;
1593                 }
1594         } else {
1595                 ret = btrfs_get_bdev_and_sb(device_path,
1596                                             FMODE_WRITE | FMODE_EXCL,
1597                                             root->fs_info->bdev_holder, 0,
1598                                             &bdev, &bh);
1599                 if (ret)
1600                         goto out;
1601                 disk_super = (struct btrfs_super_block *)bh->b_data;
1602                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1603                 dev_uuid = disk_super->dev_item.uuid;
1604                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1605                                            disk_super->fsid);
1606                 if (!device) {
1607                         ret = -ENOENT;
1608                         goto error_brelse;
1609                 }
1610         }
1611
1612         if (device->is_tgtdev_for_dev_replace) {
1613                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1614                 goto error_brelse;
1615         }
1616
1617         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1618                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1619                 goto error_brelse;
1620         }
1621
1622         if (device->writeable) {
1623                 lock_chunks(root);
1624                 list_del_init(&device->dev_alloc_list);
1625                 unlock_chunks(root);
1626                 root->fs_info->fs_devices->rw_devices--;
1627                 clear_super = true;
1628         }
1629
1630         mutex_unlock(&uuid_mutex);
1631         ret = btrfs_shrink_device(device, 0);
1632         mutex_lock(&uuid_mutex);
1633         if (ret)
1634                 goto error_undo;
1635
1636         /*
1637          * TODO: the superblock still includes this device in its num_devices
1638          * counter although write_all_supers() is not locked out. This
1639          * could give a filesystem state which requires a degraded mount.
1640          */
1641         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1642         if (ret)
1643                 goto error_undo;
1644
1645         spin_lock(&root->fs_info->free_chunk_lock);
1646         root->fs_info->free_chunk_space = device->total_bytes -
1647                 device->bytes_used;
1648         spin_unlock(&root->fs_info->free_chunk_lock);
1649
1650         device->in_fs_metadata = 0;
1651         btrfs_scrub_cancel_dev(root->fs_info, device);
1652
1653         /*
1654          * the device list mutex makes sure that we don't change
1655          * the device list while someone else is writing out all
1656          * the device supers. Whoever is writing all supers, should
1657          * lock the device list mutex before getting the number of
1658          * devices in the super block (super_copy). Conversely,
1659          * whoever updates the number of devices in the super block
1660          * (super_copy) should hold the device list mutex.
1661          */
1662
1663         cur_devices = device->fs_devices;
1664         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1665         list_del_rcu(&device->dev_list);
1666
1667         device->fs_devices->num_devices--;
1668         device->fs_devices->total_devices--;
1669
1670         if (device->missing)
1671                 root->fs_info->fs_devices->missing_devices--;
1672
1673         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1674                                  struct btrfs_device, dev_list);
1675         if (device->bdev == root->fs_info->sb->s_bdev)
1676                 root->fs_info->sb->s_bdev = next_device->bdev;
1677         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1678                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1679
1680         if (device->bdev)
1681                 device->fs_devices->open_devices--;
1682
1683         call_rcu(&device->rcu, free_device);
1684
1685         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1686         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1687         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1688
1689         if (cur_devices->open_devices == 0) {
1690                 struct btrfs_fs_devices *fs_devices;
1691                 fs_devices = root->fs_info->fs_devices;
1692                 while (fs_devices) {
1693                         if (fs_devices->seed == cur_devices) {
1694                                 fs_devices->seed = cur_devices->seed;
1695                                 break;
1696                         }
1697                         fs_devices = fs_devices->seed;
1698                 }
1699                 cur_devices->seed = NULL;
1700                 lock_chunks(root);
1701                 __btrfs_close_devices(cur_devices);
1702                 unlock_chunks(root);
1703                 free_fs_devices(cur_devices);
1704         }
1705
1706         root->fs_info->num_tolerated_disk_barrier_failures =
1707                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1708
1709         /*
1710          * at this point, the device is zero sized.  We want to
1711          * remove it from the devices list and zero out the old super
1712          */
1713         if (clear_super && disk_super) {
1714                 u64 bytenr;
1715                 int i;
1716
1717                 /* make sure this device isn't detected as part of
1718                  * the FS anymore
1719                  */
1720                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1721                 set_buffer_dirty(bh);
1722                 sync_dirty_buffer(bh);
1723
1724                 /* clear the mirror copies of super block on the disk
1725                  * being removed, 0th copy is been taken care above and
1726                  * the below would take of the rest
1727                  */
1728                 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1729                         bytenr = btrfs_sb_offset(i);
1730                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1731                                         i_size_read(bdev->bd_inode))
1732                                 break;
1733
1734                         brelse(bh);
1735                         bh = __bread(bdev, bytenr / 4096,
1736                                         BTRFS_SUPER_INFO_SIZE);
1737                         if (!bh)
1738                                 continue;
1739
1740                         disk_super = (struct btrfs_super_block *)bh->b_data;
1741
1742                         if (btrfs_super_bytenr(disk_super) != bytenr ||
1743                                 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1744                                 continue;
1745                         }
1746                         memset(&disk_super->magic, 0,
1747                                                 sizeof(disk_super->magic));
1748                         set_buffer_dirty(bh);
1749                         sync_dirty_buffer(bh);
1750                 }
1751         }
1752
1753         ret = 0;
1754
1755         if (bdev) {
1756                 /* Notify udev that device has changed */
1757                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1758
1759                 /* Update ctime/mtime for device path for libblkid */
1760                 update_dev_time(device_path);
1761         }
1762
1763 error_brelse:
1764         brelse(bh);
1765         if (bdev)
1766                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1767 out:
1768         mutex_unlock(&uuid_mutex);
1769         return ret;
1770 error_undo:
1771         if (device->writeable) {
1772                 lock_chunks(root);
1773                 list_add(&device->dev_alloc_list,
1774                          &root->fs_info->fs_devices->alloc_list);
1775                 unlock_chunks(root);
1776                 root->fs_info->fs_devices->rw_devices++;
1777         }
1778         goto error_brelse;
1779 }
1780
1781 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1782                                  struct btrfs_device *srcdev)
1783 {
1784         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1785
1786         list_del_rcu(&srcdev->dev_list);
1787         list_del_rcu(&srcdev->dev_alloc_list);
1788         fs_info->fs_devices->num_devices--;
1789         if (srcdev->missing) {
1790                 fs_info->fs_devices->missing_devices--;
1791                 fs_info->fs_devices->rw_devices++;
1792         }
1793         if (srcdev->can_discard)
1794                 fs_info->fs_devices->num_can_discard--;
1795         if (srcdev->bdev) {
1796                 fs_info->fs_devices->open_devices--;
1797
1798                 /* zero out the old super */
1799                 btrfs_scratch_superblock(srcdev);
1800         }
1801
1802         call_rcu(&srcdev->rcu, free_device);
1803 }
1804
1805 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1806                                       struct btrfs_device *tgtdev)
1807 {
1808         struct btrfs_device *next_device;
1809
1810         WARN_ON(!tgtdev);
1811         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1812         if (tgtdev->bdev) {
1813                 btrfs_scratch_superblock(tgtdev);
1814                 fs_info->fs_devices->open_devices--;
1815         }
1816         fs_info->fs_devices->num_devices--;
1817         if (tgtdev->can_discard)
1818                 fs_info->fs_devices->num_can_discard++;
1819
1820         next_device = list_entry(fs_info->fs_devices->devices.next,
1821                                  struct btrfs_device, dev_list);
1822         if (tgtdev->bdev == fs_info->sb->s_bdev)
1823                 fs_info->sb->s_bdev = next_device->bdev;
1824         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1825                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1826         list_del_rcu(&tgtdev->dev_list);
1827
1828         call_rcu(&tgtdev->rcu, free_device);
1829
1830         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1831 }
1832
1833 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1834                                      struct btrfs_device **device)
1835 {
1836         int ret = 0;
1837         struct btrfs_super_block *disk_super;
1838         u64 devid;
1839         u8 *dev_uuid;
1840         struct block_device *bdev;
1841         struct buffer_head *bh;
1842
1843         *device = NULL;
1844         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1845                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1846         if (ret)
1847                 return ret;
1848         disk_super = (struct btrfs_super_block *)bh->b_data;
1849         devid = btrfs_stack_device_id(&disk_super->dev_item);
1850         dev_uuid = disk_super->dev_item.uuid;
1851         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1852                                     disk_super->fsid);
1853         brelse(bh);
1854         if (!*device)
1855                 ret = -ENOENT;
1856         blkdev_put(bdev, FMODE_READ);
1857         return ret;
1858 }
1859
1860 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1861                                          char *device_path,
1862                                          struct btrfs_device **device)
1863 {
1864         *device = NULL;
1865         if (strcmp(device_path, "missing") == 0) {
1866                 struct list_head *devices;
1867                 struct btrfs_device *tmp;
1868
1869                 devices = &root->fs_info->fs_devices->devices;
1870                 /*
1871                  * It is safe to read the devices since the volume_mutex
1872                  * is held by the caller.
1873                  */
1874                 list_for_each_entry(tmp, devices, dev_list) {
1875                         if (tmp->in_fs_metadata && !tmp->bdev) {
1876                                 *device = tmp;
1877                                 break;
1878                         }
1879                 }
1880
1881                 if (!*device) {
1882                         btrfs_err(root->fs_info, "no missing device found");
1883                         return -ENOENT;
1884                 }
1885
1886                 return 0;
1887         } else {
1888                 return btrfs_find_device_by_path(root, device_path, device);
1889         }
1890 }
1891
1892 /*
1893  * does all the dirty work required for changing file system's UUID.
1894  */
1895 static int btrfs_prepare_sprout(struct btrfs_root *root)
1896 {
1897         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1898         struct btrfs_fs_devices *old_devices;
1899         struct btrfs_fs_devices *seed_devices;
1900         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1901         struct btrfs_device *device;
1902         u64 super_flags;
1903
1904         BUG_ON(!mutex_is_locked(&uuid_mutex));
1905         if (!fs_devices->seeding)
1906                 return -EINVAL;
1907
1908         seed_devices = __alloc_fs_devices();
1909         if (IS_ERR(seed_devices))
1910                 return PTR_ERR(seed_devices);
1911
1912         old_devices = clone_fs_devices(fs_devices);
1913         if (IS_ERR(old_devices)) {
1914                 kfree(seed_devices);
1915                 return PTR_ERR(old_devices);
1916         }
1917
1918         list_add(&old_devices->list, &fs_uuids);
1919
1920         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1921         seed_devices->opened = 1;
1922         INIT_LIST_HEAD(&seed_devices->devices);
1923         INIT_LIST_HEAD(&seed_devices->alloc_list);
1924         mutex_init(&seed_devices->device_list_mutex);
1925
1926         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1927         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1928                               synchronize_rcu);
1929
1930         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1931         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1932                 device->fs_devices = seed_devices;
1933         }
1934
1935         fs_devices->seeding = 0;
1936         fs_devices->num_devices = 0;
1937         fs_devices->open_devices = 0;
1938         fs_devices->seed = seed_devices;
1939
1940         generate_random_uuid(fs_devices->fsid);
1941         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1942         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1943         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1944
1945         super_flags = btrfs_super_flags(disk_super) &
1946                       ~BTRFS_SUPER_FLAG_SEEDING;
1947         btrfs_set_super_flags(disk_super, super_flags);
1948
1949         return 0;
1950 }
1951
1952 /*
1953  * strore the expected generation for seed devices in device items.
1954  */
1955 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1956                                struct btrfs_root *root)
1957 {
1958         struct btrfs_path *path;
1959         struct extent_buffer *leaf;
1960         struct btrfs_dev_item *dev_item;
1961         struct btrfs_device *device;
1962         struct btrfs_key key;
1963         u8 fs_uuid[BTRFS_UUID_SIZE];
1964         u8 dev_uuid[BTRFS_UUID_SIZE];
1965         u64 devid;
1966         int ret;
1967
1968         path = btrfs_alloc_path();
1969         if (!path)
1970                 return -ENOMEM;
1971
1972         root = root->fs_info->chunk_root;
1973         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1974         key.offset = 0;
1975         key.type = BTRFS_DEV_ITEM_KEY;
1976
1977         while (1) {
1978                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1979                 if (ret < 0)
1980                         goto error;
1981
1982                 leaf = path->nodes[0];
1983 next_slot:
1984                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1985                         ret = btrfs_next_leaf(root, path);
1986                         if (ret > 0)
1987                                 break;
1988                         if (ret < 0)
1989                                 goto error;
1990                         leaf = path->nodes[0];
1991                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1992                         btrfs_release_path(path);
1993                         continue;
1994                 }
1995
1996                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1997                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1998                     key.type != BTRFS_DEV_ITEM_KEY)
1999                         break;
2000
2001                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2002                                           struct btrfs_dev_item);
2003                 devid = btrfs_device_id(leaf, dev_item);
2004                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2005                                    BTRFS_UUID_SIZE);
2006                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2007                                    BTRFS_UUID_SIZE);
2008                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2009                                            fs_uuid);
2010                 BUG_ON(!device); /* Logic error */
2011
2012                 if (device->fs_devices->seeding) {
2013                         btrfs_set_device_generation(leaf, dev_item,
2014                                                     device->generation);
2015                         btrfs_mark_buffer_dirty(leaf);
2016                 }
2017
2018                 path->slots[0]++;
2019                 goto next_slot;
2020         }
2021         ret = 0;
2022 error:
2023         btrfs_free_path(path);
2024         return ret;
2025 }
2026
2027 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2028 {
2029         struct request_queue *q;
2030         struct btrfs_trans_handle *trans;
2031         struct btrfs_device *device;
2032         struct block_device *bdev;
2033         struct list_head *devices;
2034         struct super_block *sb = root->fs_info->sb;
2035         struct rcu_string *name;
2036         u64 total_bytes;
2037         int seeding_dev = 0;
2038         int ret = 0;
2039
2040         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2041                 return -EROFS;
2042
2043         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2044                                   root->fs_info->bdev_holder);
2045         if (IS_ERR(bdev))
2046                 return PTR_ERR(bdev);
2047
2048         if (root->fs_info->fs_devices->seeding) {
2049                 seeding_dev = 1;
2050                 down_write(&sb->s_umount);
2051                 mutex_lock(&uuid_mutex);
2052         }
2053
2054         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2055
2056         devices = &root->fs_info->fs_devices->devices;
2057
2058         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2059         list_for_each_entry(device, devices, dev_list) {
2060                 if (device->bdev == bdev) {
2061                         ret = -EEXIST;
2062                         mutex_unlock(
2063                                 &root->fs_info->fs_devices->device_list_mutex);
2064                         goto error;
2065                 }
2066         }
2067         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2068
2069         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2070         if (IS_ERR(device)) {
2071                 /* we can safely leave the fs_devices entry around */
2072                 ret = PTR_ERR(device);
2073                 goto error;
2074         }
2075
2076         name = rcu_string_strdup(device_path, GFP_NOFS);
2077         if (!name) {
2078                 kfree(device);
2079                 ret = -ENOMEM;
2080                 goto error;
2081         }
2082         rcu_assign_pointer(device->name, name);
2083
2084         trans = btrfs_start_transaction(root, 0);
2085         if (IS_ERR(trans)) {
2086                 rcu_string_free(device->name);
2087                 kfree(device);
2088                 ret = PTR_ERR(trans);
2089                 goto error;
2090         }
2091
2092         lock_chunks(root);
2093
2094         q = bdev_get_queue(bdev);
2095         if (blk_queue_discard(q))
2096                 device->can_discard = 1;
2097         device->writeable = 1;
2098         device->generation = trans->transid;
2099         device->io_width = root->sectorsize;
2100         device->io_align = root->sectorsize;
2101         device->sector_size = root->sectorsize;
2102         device->total_bytes = i_size_read(bdev->bd_inode);
2103         device->disk_total_bytes = device->total_bytes;
2104         device->dev_root = root->fs_info->dev_root;
2105         device->bdev = bdev;
2106         device->in_fs_metadata = 1;
2107         device->is_tgtdev_for_dev_replace = 0;
2108         device->mode = FMODE_EXCL;
2109         device->dev_stats_valid = 1;
2110         set_blocksize(device->bdev, 4096);
2111
2112         if (seeding_dev) {
2113                 sb->s_flags &= ~MS_RDONLY;
2114                 ret = btrfs_prepare_sprout(root);
2115                 BUG_ON(ret); /* -ENOMEM */
2116         }
2117
2118         device->fs_devices = root->fs_info->fs_devices;
2119
2120         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2121         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2122         list_add(&device->dev_alloc_list,
2123                  &root->fs_info->fs_devices->alloc_list);
2124         root->fs_info->fs_devices->num_devices++;
2125         root->fs_info->fs_devices->open_devices++;
2126         root->fs_info->fs_devices->rw_devices++;
2127         root->fs_info->fs_devices->total_devices++;
2128         if (device->can_discard)
2129                 root->fs_info->fs_devices->num_can_discard++;
2130         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2131
2132         spin_lock(&root->fs_info->free_chunk_lock);
2133         root->fs_info->free_chunk_space += device->total_bytes;
2134         spin_unlock(&root->fs_info->free_chunk_lock);
2135
2136         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2137                 root->fs_info->fs_devices->rotating = 1;
2138
2139         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2140         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2141                                     total_bytes + device->total_bytes);
2142
2143         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2144         btrfs_set_super_num_devices(root->fs_info->super_copy,
2145                                     total_bytes + 1);
2146         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2147
2148         if (seeding_dev) {
2149                 ret = init_first_rw_device(trans, root, device);
2150                 if (ret) {
2151                         btrfs_abort_transaction(trans, root, ret);
2152                         goto error_trans;
2153                 }
2154                 ret = btrfs_finish_sprout(trans, root);
2155                 if (ret) {
2156                         btrfs_abort_transaction(trans, root, ret);
2157                         goto error_trans;
2158                 }
2159         } else {
2160                 ret = btrfs_add_device(trans, root, device);
2161                 if (ret) {
2162                         btrfs_abort_transaction(trans, root, ret);
2163                         goto error_trans;
2164                 }
2165         }
2166
2167         /*
2168          * we've got more storage, clear any full flags on the space
2169          * infos
2170          */
2171         btrfs_clear_space_info_full(root->fs_info);
2172
2173         unlock_chunks(root);
2174         root->fs_info->num_tolerated_disk_barrier_failures =
2175                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2176         ret = btrfs_commit_transaction(trans, root);
2177
2178         if (seeding_dev) {
2179                 mutex_unlock(&uuid_mutex);
2180                 up_write(&sb->s_umount);
2181
2182                 if (ret) /* transaction commit */
2183                         return ret;
2184
2185                 ret = btrfs_relocate_sys_chunks(root);
2186                 if (ret < 0)
2187                         btrfs_error(root->fs_info, ret,
2188                                     "Failed to relocate sys chunks after "
2189                                     "device initialization. This can be fixed "
2190                                     "using the \"btrfs balance\" command.");
2191                 trans = btrfs_attach_transaction(root);
2192                 if (IS_ERR(trans)) {
2193                         if (PTR_ERR(trans) == -ENOENT)
2194                                 return 0;
2195                         return PTR_ERR(trans);
2196                 }
2197                 ret = btrfs_commit_transaction(trans, root);
2198         }
2199
2200         /* Update ctime/mtime for libblkid */
2201         update_dev_time(device_path);
2202         return ret;
2203
2204 error_trans:
2205         unlock_chunks(root);
2206         btrfs_end_transaction(trans, root);
2207         rcu_string_free(device->name);
2208         kfree(device);
2209 error:
2210         blkdev_put(bdev, FMODE_EXCL);
2211         if (seeding_dev) {
2212                 mutex_unlock(&uuid_mutex);
2213                 up_write(&sb->s_umount);
2214         }
2215         return ret;
2216 }
2217
2218 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2219                                   struct btrfs_device **device_out)
2220 {
2221         struct request_queue *q;
2222         struct btrfs_device *device;
2223         struct block_device *bdev;
2224         struct btrfs_fs_info *fs_info = root->fs_info;
2225         struct list_head *devices;
2226         struct rcu_string *name;
2227         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2228         int ret = 0;
2229
2230         *device_out = NULL;
2231         if (fs_info->fs_devices->seeding)
2232                 return -EINVAL;
2233
2234         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2235                                   fs_info->bdev_holder);
2236         if (IS_ERR(bdev))
2237                 return PTR_ERR(bdev);
2238
2239         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2240
2241         devices = &fs_info->fs_devices->devices;
2242         list_for_each_entry(device, devices, dev_list) {
2243                 if (device->bdev == bdev) {
2244                         ret = -EEXIST;
2245                         goto error;
2246                 }
2247         }
2248
2249         device = btrfs_alloc_device(NULL, &devid, NULL);
2250         if (IS_ERR(device)) {
2251                 ret = PTR_ERR(device);
2252                 goto error;
2253         }
2254
2255         name = rcu_string_strdup(device_path, GFP_NOFS);
2256         if (!name) {
2257                 kfree(device);
2258                 ret = -ENOMEM;
2259                 goto error;
2260         }
2261         rcu_assign_pointer(device->name, name);
2262
2263         q = bdev_get_queue(bdev);
2264         if (blk_queue_discard(q))
2265                 device->can_discard = 1;
2266         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2267         device->writeable = 1;
2268         device->generation = 0;
2269         device->io_width = root->sectorsize;
2270         device->io_align = root->sectorsize;
2271         device->sector_size = root->sectorsize;
2272         device->total_bytes = i_size_read(bdev->bd_inode);
2273         device->disk_total_bytes = device->total_bytes;
2274         device->dev_root = fs_info->dev_root;
2275         device->bdev = bdev;
2276         device->in_fs_metadata = 1;
2277         device->is_tgtdev_for_dev_replace = 1;
2278         device->mode = FMODE_EXCL;
2279         device->dev_stats_valid = 1;
2280         set_blocksize(device->bdev, 4096);
2281         device->fs_devices = fs_info->fs_devices;
2282         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2283         fs_info->fs_devices->num_devices++;
2284         fs_info->fs_devices->open_devices++;
2285         if (device->can_discard)
2286                 fs_info->fs_devices->num_can_discard++;
2287         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2288
2289         *device_out = device;
2290         return ret;
2291
2292 error:
2293         blkdev_put(bdev, FMODE_EXCL);
2294         return ret;
2295 }
2296
2297 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2298                                               struct btrfs_device *tgtdev)
2299 {
2300         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2301         tgtdev->io_width = fs_info->dev_root->sectorsize;
2302         tgtdev->io_align = fs_info->dev_root->sectorsize;
2303         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2304         tgtdev->dev_root = fs_info->dev_root;
2305         tgtdev->in_fs_metadata = 1;
2306 }
2307
2308 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2309                                         struct btrfs_device *device)
2310 {
2311         int ret;
2312         struct btrfs_path *path;
2313         struct btrfs_root *root;
2314         struct btrfs_dev_item *dev_item;
2315         struct extent_buffer *leaf;
2316         struct btrfs_key key;
2317
2318         root = device->dev_root->fs_info->chunk_root;
2319
2320         path = btrfs_alloc_path();
2321         if (!path)
2322                 return -ENOMEM;
2323
2324         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2325         key.type = BTRFS_DEV_ITEM_KEY;
2326         key.offset = device->devid;
2327
2328         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2329         if (ret < 0)
2330                 goto out;
2331
2332         if (ret > 0) {
2333                 ret = -ENOENT;
2334                 goto out;
2335         }
2336
2337         leaf = path->nodes[0];
2338         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2339
2340         btrfs_set_device_id(leaf, dev_item, device->devid);
2341         btrfs_set_device_type(leaf, dev_item, device->type);
2342         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2343         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2344         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2345         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2346         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2347         btrfs_mark_buffer_dirty(leaf);
2348
2349 out:
2350         btrfs_free_path(path);
2351         return ret;
2352 }
2353
2354 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2355                       struct btrfs_device *device, u64 new_size)
2356 {
2357         struct btrfs_super_block *super_copy =
2358                 device->dev_root->fs_info->super_copy;
2359         u64 old_total = btrfs_super_total_bytes(super_copy);
2360         u64 diff = new_size - device->total_bytes;
2361
2362         if (!device->writeable)
2363                 return -EACCES;
2364         if (new_size <= device->total_bytes ||
2365             device->is_tgtdev_for_dev_replace)
2366                 return -EINVAL;
2367
2368         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2369         device->fs_devices->total_rw_bytes += diff;
2370
2371         device->total_bytes = new_size;
2372         device->disk_total_bytes = new_size;
2373         btrfs_clear_space_info_full(device->dev_root->fs_info);
2374
2375         return btrfs_update_device(trans, device);
2376 }
2377
2378 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2379                       struct btrfs_device *device, u64 new_size)
2380 {
2381         int ret;
2382         lock_chunks(device->dev_root);
2383         ret = __btrfs_grow_device(trans, device, new_size);
2384         unlock_chunks(device->dev_root);
2385         return ret;
2386 }
2387
2388 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2389                             struct btrfs_root *root,
2390                             u64 chunk_tree, u64 chunk_objectid,
2391                             u64 chunk_offset)
2392 {
2393         int ret;
2394         struct btrfs_path *path;
2395         struct btrfs_key key;
2396
2397         root = root->fs_info->chunk_root;
2398         path = btrfs_alloc_path();
2399         if (!path)
2400                 return -ENOMEM;
2401
2402         key.objectid = chunk_objectid;
2403         key.offset = chunk_offset;
2404         key.type = BTRFS_CHUNK_ITEM_KEY;
2405
2406         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2407         if (ret < 0)
2408                 goto out;
2409         else if (ret > 0) { /* Logic error or corruption */
2410                 btrfs_error(root->fs_info, -ENOENT,
2411                             "Failed lookup while freeing chunk.");
2412                 ret = -ENOENT;
2413                 goto out;
2414         }
2415
2416         ret = btrfs_del_item(trans, root, path);
2417         if (ret < 0)
2418                 btrfs_error(root->fs_info, ret,
2419                             "Failed to delete chunk item.");
2420 out:
2421         btrfs_free_path(path);
2422         return ret;
2423 }
2424
2425 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2426                         chunk_offset)
2427 {
2428         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2429         struct btrfs_disk_key *disk_key;
2430         struct btrfs_chunk *chunk;
2431         u8 *ptr;
2432         int ret = 0;
2433         u32 num_stripes;
2434         u32 array_size;
2435         u32 len = 0;
2436         u32 cur;
2437         struct btrfs_key key;
2438
2439         array_size = btrfs_super_sys_array_size(super_copy);
2440
2441         ptr = super_copy->sys_chunk_array;
2442         cur = 0;
2443
2444         while (cur < array_size) {
2445                 disk_key = (struct btrfs_disk_key *)ptr;
2446                 btrfs_disk_key_to_cpu(&key, disk_key);
2447
2448                 len = sizeof(*disk_key);
2449
2450                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2451                         chunk = (struct btrfs_chunk *)(ptr + len);
2452                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2453                         len += btrfs_chunk_item_size(num_stripes);
2454                 } else {
2455                         ret = -EIO;
2456                         break;
2457                 }
2458                 if (key.objectid == chunk_objectid &&
2459                     key.offset == chunk_offset) {
2460                         memmove(ptr, ptr + len, array_size - (cur + len));
2461                         array_size -= len;
2462                         btrfs_set_super_sys_array_size(super_copy, array_size);
2463                 } else {
2464                         ptr += len;
2465                         cur += len;
2466                 }
2467         }
2468         return ret;
2469 }
2470
2471 static int btrfs_relocate_chunk(struct btrfs_root *root,
2472                          u64 chunk_tree, u64 chunk_objectid,
2473                          u64 chunk_offset)
2474 {
2475         struct extent_map_tree *em_tree;
2476         struct btrfs_root *extent_root;
2477         struct btrfs_trans_handle *trans;
2478         struct extent_map *em;
2479         struct map_lookup *map;
2480         int ret;
2481         int i;
2482
2483         root = root->fs_info->chunk_root;
2484         extent_root = root->fs_info->extent_root;
2485         em_tree = &root->fs_info->mapping_tree.map_tree;
2486
2487         ret = btrfs_can_relocate(extent_root, chunk_offset);
2488         if (ret)
2489                 return -ENOSPC;
2490
2491         /* step one, relocate all the extents inside this chunk */
2492         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2493         if (ret)
2494                 return ret;
2495
2496         trans = btrfs_start_transaction(root, 0);
2497         if (IS_ERR(trans)) {
2498                 ret = PTR_ERR(trans);
2499                 btrfs_std_error(root->fs_info, ret);
2500                 return ret;
2501         }
2502
2503         lock_chunks(root);
2504
2505         /*
2506          * step two, delete the device extents and the
2507          * chunk tree entries
2508          */
2509         read_lock(&em_tree->lock);
2510         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2511         read_unlock(&em_tree->lock);
2512
2513         BUG_ON(!em || em->start > chunk_offset ||
2514                em->start + em->len < chunk_offset);
2515         map = (struct map_lookup *)em->bdev;
2516
2517         for (i = 0; i < map->num_stripes; i++) {
2518                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2519                                             map->stripes[i].physical);
2520                 BUG_ON(ret);
2521
2522                 if (map->stripes[i].dev) {
2523                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2524                         BUG_ON(ret);
2525                 }
2526         }
2527         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2528                                chunk_offset);
2529
2530         BUG_ON(ret);
2531
2532         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2533
2534         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2535                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2536                 BUG_ON(ret);
2537         }
2538
2539         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2540         BUG_ON(ret);
2541
2542         write_lock(&em_tree->lock);
2543         remove_extent_mapping(em_tree, em);
2544         write_unlock(&em_tree->lock);
2545
2546         /* once for the tree */
2547         free_extent_map(em);
2548         /* once for us */
2549         free_extent_map(em);
2550
2551         unlock_chunks(root);
2552         btrfs_end_transaction(trans, root);
2553         return 0;
2554 }
2555
2556 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2557 {
2558         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2559         struct btrfs_path *path;
2560         struct extent_buffer *leaf;
2561         struct btrfs_chunk *chunk;
2562         struct btrfs_key key;
2563         struct btrfs_key found_key;
2564         u64 chunk_tree = chunk_root->root_key.objectid;
2565         u64 chunk_type;
2566         bool retried = false;
2567         int failed = 0;
2568         int ret;
2569
2570         path = btrfs_alloc_path();
2571         if (!path)
2572                 return -ENOMEM;
2573
2574 again:
2575         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2576         key.offset = (u64)-1;
2577         key.type = BTRFS_CHUNK_ITEM_KEY;
2578
2579         while (1) {
2580                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2581                 if (ret < 0)
2582                         goto error;
2583                 BUG_ON(ret == 0); /* Corruption */
2584
2585                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2586                                           key.type);
2587                 if (ret < 0)
2588                         goto error;
2589                 if (ret > 0)
2590                         break;
2591
2592                 leaf = path->nodes[0];
2593                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2594
2595                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2596                                        struct btrfs_chunk);
2597                 chunk_type = btrfs_chunk_type(leaf, chunk);
2598                 btrfs_release_path(path);
2599
2600                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2601                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2602                                                    found_key.objectid,
2603                                                    found_key.offset);
2604                         if (ret == -ENOSPC)
2605                                 failed++;
2606                         else if (ret)
2607                                 BUG();
2608                 }
2609
2610                 if (found_key.offset == 0)
2611                         break;
2612                 key.offset = found_key.offset - 1;
2613         }
2614         ret = 0;
2615         if (failed && !retried) {
2616                 failed = 0;
2617                 retried = true;
2618                 goto again;
2619         } else if (WARN_ON(failed && retried)) {
2620                 ret = -ENOSPC;
2621         }
2622 error:
2623         btrfs_free_path(path);
2624         return ret;
2625 }
2626
2627 static int insert_balance_item(struct btrfs_root *root,
2628                                struct btrfs_balance_control *bctl)
2629 {
2630         struct btrfs_trans_handle *trans;
2631         struct btrfs_balance_item *item;
2632         struct btrfs_disk_balance_args disk_bargs;
2633         struct btrfs_path *path;
2634         struct extent_buffer *leaf;
2635         struct btrfs_key key;
2636         int ret, err;
2637
2638         path = btrfs_alloc_path();
2639         if (!path)
2640                 return -ENOMEM;
2641
2642         trans = btrfs_start_transaction(root, 0);
2643         if (IS_ERR(trans)) {
2644                 btrfs_free_path(path);
2645                 return PTR_ERR(trans);
2646         }
2647
2648         key.objectid = BTRFS_BALANCE_OBJECTID;
2649         key.type = BTRFS_BALANCE_ITEM_KEY;
2650         key.offset = 0;
2651
2652         ret = btrfs_insert_empty_item(trans, root, path, &key,
2653                                       sizeof(*item));
2654         if (ret)
2655                 goto out;
2656
2657         leaf = path->nodes[0];
2658         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2659
2660         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2661
2662         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2663         btrfs_set_balance_data(leaf, item, &disk_bargs);
2664         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2665         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2666         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2667         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2668
2669         btrfs_set_balance_flags(leaf, item, bctl->flags);
2670
2671         btrfs_mark_buffer_dirty(leaf);
2672 out:
2673         btrfs_free_path(path);
2674         err = btrfs_commit_transaction(trans, root);
2675         if (err && !ret)
2676                 ret = err;
2677         return ret;
2678 }
2679
2680 static int del_balance_item(struct btrfs_root *root)
2681 {
2682         struct btrfs_trans_handle *trans;
2683         struct btrfs_path *path;
2684         struct btrfs_key key;
2685         int ret, err;
2686
2687         path = btrfs_alloc_path();
2688         if (!path)
2689                 return -ENOMEM;
2690
2691         trans = btrfs_start_transaction(root, 0);
2692         if (IS_ERR(trans)) {
2693                 btrfs_free_path(path);
2694                 return PTR_ERR(trans);
2695         }
2696
2697         key.objectid = BTRFS_BALANCE_OBJECTID;
2698         key.type = BTRFS_BALANCE_ITEM_KEY;
2699         key.offset = 0;
2700
2701         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2702         if (ret < 0)
2703                 goto out;
2704         if (ret > 0) {
2705                 ret = -ENOENT;
2706                 goto out;
2707         }
2708
2709         ret = btrfs_del_item(trans, root, path);
2710 out:
2711         btrfs_free_path(path);
2712         err = btrfs_commit_transaction(trans, root);
2713         if (err && !ret)
2714                 ret = err;
2715         return ret;
2716 }
2717
2718 /*
2719  * This is a heuristic used to reduce the number of chunks balanced on
2720  * resume after balance was interrupted.
2721  */
2722 static void update_balance_args(struct btrfs_balance_control *bctl)
2723 {
2724         /*
2725          * Turn on soft mode for chunk types that were being converted.
2726          */
2727         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2728                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2729         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2730                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2731         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2732                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2733
2734         /*
2735          * Turn on usage filter if is not already used.  The idea is
2736          * that chunks that we have already balanced should be
2737          * reasonably full.  Don't do it for chunks that are being
2738          * converted - that will keep us from relocating unconverted
2739          * (albeit full) chunks.
2740          */
2741         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2742             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2743                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2744                 bctl->data.usage = 90;
2745         }
2746         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2747             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2748                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2749                 bctl->sys.usage = 90;
2750         }
2751         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2752             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2753                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2754                 bctl->meta.usage = 90;
2755         }
2756 }
2757
2758 /*
2759  * Should be called with both balance and volume mutexes held to
2760  * serialize other volume operations (add_dev/rm_dev/resize) with
2761  * restriper.  Same goes for unset_balance_control.
2762  */
2763 static void set_balance_control(struct btrfs_balance_control *bctl)
2764 {
2765         struct btrfs_fs_info *fs_info = bctl->fs_info;
2766
2767         BUG_ON(fs_info->balance_ctl);
2768
2769         spin_lock(&fs_info->balance_lock);
2770         fs_info->balance_ctl = bctl;
2771         spin_unlock(&fs_info->balance_lock);
2772 }
2773
2774 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2775 {
2776         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2777
2778         BUG_ON(!fs_info->balance_ctl);
2779
2780         spin_lock(&fs_info->balance_lock);
2781         fs_info->balance_ctl = NULL;
2782         spin_unlock(&fs_info->balance_lock);
2783
2784         kfree(bctl);
2785 }
2786
2787 /*
2788  * Balance filters.  Return 1 if chunk should be filtered out
2789  * (should not be balanced).
2790  */
2791 static int chunk_profiles_filter(u64 chunk_type,
2792                                  struct btrfs_balance_args *bargs)
2793 {
2794         chunk_type = chunk_to_extended(chunk_type) &
2795                                 BTRFS_EXTENDED_PROFILE_MASK;
2796
2797         if (bargs->profiles & chunk_type)
2798                 return 0;
2799
2800         return 1;
2801 }
2802
2803 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2804                               struct btrfs_balance_args *bargs)
2805 {
2806         struct btrfs_block_group_cache *cache;
2807         u64 chunk_used, user_thresh;
2808         int ret = 1;
2809
2810         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2811         chunk_used = btrfs_block_group_used(&cache->item);
2812
2813         if (bargs->usage == 0)
2814                 user_thresh = 1;
2815         else if (bargs->usage > 100)
2816                 user_thresh = cache->key.offset;
2817         else
2818                 user_thresh = div_factor_fine(cache->key.offset,
2819                                               bargs->usage);
2820
2821         if (chunk_used < user_thresh)
2822                 ret = 0;
2823
2824         btrfs_put_block_group(cache);
2825         return ret;
2826 }
2827
2828 static int chunk_devid_filter(struct extent_buffer *leaf,
2829                               struct btrfs_chunk *chunk,
2830                               struct btrfs_balance_args *bargs)
2831 {
2832         struct btrfs_stripe *stripe;
2833         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2834         int i;
2835
2836         for (i = 0; i < num_stripes; i++) {
2837                 stripe = btrfs_stripe_nr(chunk, i);
2838                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2839                         return 0;
2840         }
2841
2842         return 1;
2843 }
2844
2845 /* [pstart, pend) */
2846 static int chunk_drange_filter(struct extent_buffer *leaf,
2847                                struct btrfs_chunk *chunk,
2848                                u64 chunk_offset,
2849                                struct btrfs_balance_args *bargs)
2850 {
2851         struct btrfs_stripe *stripe;
2852         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2853         u64 stripe_offset;
2854         u64 stripe_length;
2855         int factor;
2856         int i;
2857
2858         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2859                 return 0;
2860
2861         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2862              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2863                 factor = num_stripes / 2;
2864         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2865                 factor = num_stripes - 1;
2866         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2867                 factor = num_stripes - 2;
2868         } else {
2869                 factor = num_stripes;
2870         }
2871
2872         for (i = 0; i < num_stripes; i++) {
2873                 stripe = btrfs_stripe_nr(chunk, i);
2874                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2875                         continue;
2876
2877                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2878                 stripe_length = btrfs_chunk_length(leaf, chunk);
2879                 do_div(stripe_length, factor);
2880
2881                 if (stripe_offset < bargs->pend &&
2882                     stripe_offset + stripe_length > bargs->pstart)
2883                         return 0;
2884         }
2885
2886         return 1;
2887 }
2888
2889 /* [vstart, vend) */
2890 static int chunk_vrange_filter(struct extent_buffer *leaf,
2891                                struct btrfs_chunk *chunk,
2892                                u64 chunk_offset,
2893                                struct btrfs_balance_args *bargs)
2894 {
2895         if (chunk_offset < bargs->vend &&
2896             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2897                 /* at least part of the chunk is inside this vrange */
2898                 return 0;
2899
2900         return 1;
2901 }
2902
2903 static int chunk_soft_convert_filter(u64 chunk_type,
2904                                      struct btrfs_balance_args *bargs)
2905 {
2906         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2907                 return 0;
2908
2909         chunk_type = chunk_to_extended(chunk_type) &
2910                                 BTRFS_EXTENDED_PROFILE_MASK;
2911
2912         if (bargs->target == chunk_type)
2913                 return 1;
2914
2915         return 0;
2916 }
2917
2918 static int should_balance_chunk(struct btrfs_root *root,
2919                                 struct extent_buffer *leaf,
2920                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2921 {
2922         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2923         struct btrfs_balance_args *bargs = NULL;
2924         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2925
2926         /* type filter */
2927         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2928               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2929                 return 0;
2930         }
2931
2932         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2933                 bargs = &bctl->data;
2934         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2935                 bargs = &bctl->sys;
2936         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2937                 bargs = &bctl->meta;
2938
2939         /* profiles filter */
2940         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2941             chunk_profiles_filter(chunk_type, bargs)) {
2942                 return 0;
2943         }
2944
2945         /* usage filter */
2946         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2947             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2948                 return 0;
2949         }
2950
2951         /* devid filter */
2952         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2953             chunk_devid_filter(leaf, chunk, bargs)) {
2954                 return 0;
2955         }
2956
2957         /* drange filter, makes sense only with devid filter */
2958         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2959             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2960                 return 0;
2961         }
2962
2963         /* vrange filter */
2964         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2965             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2966                 return 0;
2967         }
2968
2969         /* soft profile changing mode */
2970         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2971             chunk_soft_convert_filter(chunk_type, bargs)) {
2972                 return 0;
2973         }
2974
2975         /*
2976          * limited by count, must be the last filter
2977          */
2978         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
2979                 if (bargs->limit == 0)
2980                         return 0;
2981                 else
2982                         bargs->limit--;
2983         }
2984
2985         return 1;
2986 }
2987
2988 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2989 {
2990         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2991         struct btrfs_root *chunk_root = fs_info->chunk_root;
2992         struct btrfs_root *dev_root = fs_info->dev_root;
2993         struct list_head *devices;
2994         struct btrfs_device *device;
2995         u64 old_size;
2996         u64 size_to_free;
2997         struct btrfs_chunk *chunk;
2998         struct btrfs_path *path;
2999         struct btrfs_key key;
3000         struct btrfs_key found_key;
3001         struct btrfs_trans_handle *trans;
3002         struct extent_buffer *leaf;
3003         int slot;
3004         int ret;
3005         int enospc_errors = 0;
3006         bool counting = true;
3007         u64 limit_data = bctl->data.limit;
3008         u64 limit_meta = bctl->meta.limit;
3009         u64 limit_sys = bctl->sys.limit;
3010
3011         /* step one make some room on all the devices */
3012         devices = &fs_info->fs_devices->devices;
3013         list_for_each_entry(device, devices, dev_list) {
3014                 old_size = device->total_bytes;
3015                 size_to_free = div_factor(old_size, 1);
3016                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3017                 if (!device->writeable ||
3018                     device->total_bytes - device->bytes_used > size_to_free ||
3019                     device->is_tgtdev_for_dev_replace)
3020                         continue;
3021
3022                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3023                 if (ret == -ENOSPC)
3024                         break;
3025                 BUG_ON(ret);
3026
3027                 trans = btrfs_start_transaction(dev_root, 0);
3028                 BUG_ON(IS_ERR(trans));
3029
3030                 ret = btrfs_grow_device(trans, device, old_size);
3031                 BUG_ON(ret);
3032
3033                 btrfs_end_transaction(trans, dev_root);
3034         }
3035
3036         /* step two, relocate all the chunks */
3037         path = btrfs_alloc_path();
3038         if (!path) {
3039                 ret = -ENOMEM;
3040                 goto error;
3041         }
3042
3043         /* zero out stat counters */
3044         spin_lock(&fs_info->balance_lock);
3045         memset(&bctl->stat, 0, sizeof(bctl->stat));
3046         spin_unlock(&fs_info->balance_lock);
3047 again:
3048         if (!counting) {
3049                 bctl->data.limit = limit_data;
3050                 bctl->meta.limit = limit_meta;
3051                 bctl->sys.limit = limit_sys;
3052         }
3053         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3054         key.offset = (u64)-1;
3055         key.type = BTRFS_CHUNK_ITEM_KEY;
3056
3057         while (1) {
3058                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3059                     atomic_read(&fs_info->balance_cancel_req)) {
3060                         ret = -ECANCELED;
3061                         goto error;
3062                 }
3063
3064                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3065                 if (ret < 0)
3066                         goto error;
3067
3068                 /*
3069                  * this shouldn't happen, it means the last relocate
3070                  * failed
3071                  */
3072                 if (ret == 0)
3073                         BUG(); /* FIXME break ? */
3074
3075                 ret = btrfs_previous_item(chunk_root, path, 0,
3076                                           BTRFS_CHUNK_ITEM_KEY);
3077                 if (ret) {
3078                         ret = 0;
3079                         break;
3080                 }
3081
3082                 leaf = path->nodes[0];
3083                 slot = path->slots[0];
3084                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3085
3086                 if (found_key.objectid != key.objectid)
3087                         break;
3088
3089                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3090
3091                 if (!counting) {
3092                         spin_lock(&fs_info->balance_lock);
3093                         bctl->stat.considered++;
3094                         spin_unlock(&fs_info->balance_lock);
3095                 }
3096
3097                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3098                                            found_key.offset);
3099                 btrfs_release_path(path);
3100                 if (!ret)
3101                         goto loop;
3102
3103                 if (counting) {
3104                         spin_lock(&fs_info->balance_lock);
3105                         bctl->stat.expected++;
3106                         spin_unlock(&fs_info->balance_lock);
3107                         goto loop;
3108                 }
3109
3110                 ret = btrfs_relocate_chunk(chunk_root,
3111                                            chunk_root->root_key.objectid,
3112                                            found_key.objectid,
3113                                            found_key.offset);
3114                 if (ret && ret != -ENOSPC)
3115                         goto error;
3116                 if (ret == -ENOSPC) {
3117                         enospc_errors++;
3118                 } else {
3119                         spin_lock(&fs_info->balance_lock);
3120                         bctl->stat.completed++;
3121                         spin_unlock(&fs_info->balance_lock);
3122                 }
3123 loop:
3124                 if (found_key.offset == 0)
3125                         break;
3126                 key.offset = found_key.offset - 1;
3127         }
3128
3129         if (counting) {
3130                 btrfs_release_path(path);
3131                 counting = false;
3132                 goto again;
3133         }
3134 error:
3135         btrfs_free_path(path);
3136         if (enospc_errors) {
3137                 btrfs_info(fs_info, "%d enospc errors during balance",
3138                        enospc_errors);
3139                 if (!ret)
3140                         ret = -ENOSPC;
3141         }
3142
3143         return ret;
3144 }
3145
3146 /**
3147  * alloc_profile_is_valid - see if a given profile is valid and reduced
3148  * @flags: profile to validate
3149  * @extended: if true @flags is treated as an extended profile
3150  */
3151 static int alloc_profile_is_valid(u64 flags, int extended)
3152 {
3153         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3154                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3155
3156         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3157
3158         /* 1) check that all other bits are zeroed */
3159         if (flags & ~mask)
3160                 return 0;
3161
3162         /* 2) see if profile is reduced */
3163         if (flags == 0)
3164                 return !extended; /* "0" is valid for usual profiles */
3165
3166         /* true if exactly one bit set */
3167         return (flags & (flags - 1)) == 0;
3168 }
3169
3170 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3171 {
3172         /* cancel requested || normal exit path */
3173         return atomic_read(&fs_info->balance_cancel_req) ||
3174                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3175                  atomic_read(&fs_info->balance_cancel_req) == 0);
3176 }
3177
3178 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3179 {
3180         int ret;
3181
3182         unset_balance_control(fs_info);
3183         ret = del_balance_item(fs_info->tree_root);
3184         if (ret)
3185                 btrfs_std_error(fs_info, ret);
3186
3187         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3188 }
3189
3190 /*
3191  * Should be called with both balance and volume mutexes held
3192  */
3193 int btrfs_balance(struct btrfs_balance_control *bctl,
3194                   struct btrfs_ioctl_balance_args *bargs)
3195 {
3196         struct btrfs_fs_info *fs_info = bctl->fs_info;
3197         u64 allowed;
3198         int mixed = 0;
3199         int ret;
3200         u64 num_devices;
3201         unsigned seq;
3202
3203         if (btrfs_fs_closing(fs_info) ||
3204             atomic_read(&fs_info->balance_pause_req) ||
3205             atomic_read(&fs_info->balance_cancel_req)) {
3206                 ret = -EINVAL;
3207                 goto out;
3208         }
3209
3210         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3211         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3212                 mixed = 1;
3213
3214         /*
3215          * In case of mixed groups both data and meta should be picked,
3216          * and identical options should be given for both of them.
3217          */
3218         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3219         if (mixed && (bctl->flags & allowed)) {
3220                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3221                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3222                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3223                         btrfs_err(fs_info, "with mixed groups data and "
3224                                    "metadata balance options must be the same");
3225                         ret = -EINVAL;
3226                         goto out;
3227                 }
3228         }
3229
3230         num_devices = fs_info->fs_devices->num_devices;
3231         btrfs_dev_replace_lock(&fs_info->dev_replace);
3232         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3233                 BUG_ON(num_devices < 1);
3234                 num_devices--;
3235         }
3236         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3237         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3238         if (num_devices == 1)
3239                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3240         else if (num_devices > 1)
3241                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3242         if (num_devices > 2)
3243                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3244         if (num_devices > 3)
3245                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3246                             BTRFS_BLOCK_GROUP_RAID6);
3247         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3248             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3249              (bctl->data.target & ~allowed))) {
3250                 btrfs_err(fs_info, "unable to start balance with target "
3251                            "data profile %llu",
3252                        bctl->data.target);
3253                 ret = -EINVAL;
3254                 goto out;
3255         }
3256         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3257             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3258              (bctl->meta.target & ~allowed))) {
3259                 btrfs_err(fs_info,
3260                            "unable to start balance with target metadata profile %llu",
3261                        bctl->meta.target);
3262                 ret = -EINVAL;
3263                 goto out;
3264         }
3265         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3266             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3267              (bctl->sys.target & ~allowed))) {
3268                 btrfs_err(fs_info,
3269                            "unable to start balance with target system profile %llu",
3270                        bctl->sys.target);
3271                 ret = -EINVAL;
3272                 goto out;
3273         }
3274
3275         /* allow dup'ed data chunks only in mixed mode */
3276         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3277             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3278                 btrfs_err(fs_info, "dup for data is not allowed");
3279                 ret = -EINVAL;
3280                 goto out;
3281         }
3282
3283         /* allow to reduce meta or sys integrity only if force set */
3284         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3285                         BTRFS_BLOCK_GROUP_RAID10 |
3286                         BTRFS_BLOCK_GROUP_RAID5 |
3287                         BTRFS_BLOCK_GROUP_RAID6;
3288         do {
3289                 seq = read_seqbegin(&fs_info->profiles_lock);
3290
3291                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3292                      (fs_info->avail_system_alloc_bits & allowed) &&
3293                      !(bctl->sys.target & allowed)) ||
3294                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3295                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3296                      !(bctl->meta.target & allowed))) {
3297                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3298                                 btrfs_info(fs_info, "force reducing metadata integrity");
3299                         } else {
3300                                 btrfs_err(fs_info, "balance will reduce metadata "
3301                                            "integrity, use force if you want this");
3302                                 ret = -EINVAL;
3303                                 goto out;
3304                         }
3305                 }
3306         } while (read_seqretry(&fs_info->profiles_lock, seq));
3307
3308         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3309                 int num_tolerated_disk_barrier_failures;
3310                 u64 target = bctl->sys.target;
3311
3312                 num_tolerated_disk_barrier_failures =
3313                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3314                 if (num_tolerated_disk_barrier_failures > 0 &&
3315                     (target &
3316                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3317                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3318                         num_tolerated_disk_barrier_failures = 0;
3319                 else if (num_tolerated_disk_barrier_failures > 1 &&
3320                          (target &
3321                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3322                         num_tolerated_disk_barrier_failures = 1;
3323
3324                 fs_info->num_tolerated_disk_barrier_failures =
3325                         num_tolerated_disk_barrier_failures;
3326         }
3327
3328         ret = insert_balance_item(fs_info->tree_root, bctl);
3329         if (ret && ret != -EEXIST)
3330                 goto out;
3331
3332         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3333                 BUG_ON(ret == -EEXIST);
3334                 set_balance_control(bctl);
3335         } else {
3336                 BUG_ON(ret != -EEXIST);
3337                 spin_lock(&fs_info->balance_lock);
3338                 update_balance_args(bctl);
3339                 spin_unlock(&fs_info->balance_lock);
3340         }
3341
3342         atomic_inc(&fs_info->balance_running);
3343         mutex_unlock(&fs_info->balance_mutex);
3344
3345         ret = __btrfs_balance(fs_info);
3346
3347         mutex_lock(&fs_info->balance_mutex);
3348         atomic_dec(&fs_info->balance_running);
3349
3350         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3351                 fs_info->num_tolerated_disk_barrier_failures =
3352                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3353         }
3354
3355         if (bargs) {
3356                 memset(bargs, 0, sizeof(*bargs));
3357                 update_ioctl_balance_args(fs_info, 0, bargs);
3358         }
3359
3360         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3361             balance_need_close(fs_info)) {
3362                 __cancel_balance(fs_info);
3363         }
3364
3365         wake_up(&fs_info->balance_wait_q);
3366
3367         return ret;
3368 out:
3369         if (bctl->flags & BTRFS_BALANCE_RESUME)
3370                 __cancel_balance(fs_info);
3371         else {
3372                 kfree(bctl);
3373                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3374         }
3375         return ret;
3376 }
3377
3378 static int balance_kthread(void *data)
3379 {
3380         struct btrfs_fs_info *fs_info = data;
3381         int ret = 0;
3382
3383         mutex_lock(&fs_info->volume_mutex);
3384         mutex_lock(&fs_info->balance_mutex);
3385
3386         if (fs_info->balance_ctl) {
3387                 btrfs_info(fs_info, "continuing balance");
3388                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3389         }
3390
3391         mutex_unlock(&fs_info->balance_mutex);
3392         mutex_unlock(&fs_info->volume_mutex);
3393
3394         return ret;
3395 }
3396
3397 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3398 {
3399         struct task_struct *tsk;
3400
3401         spin_lock(&fs_info->balance_lock);
3402         if (!fs_info->balance_ctl) {
3403                 spin_unlock(&fs_info->balance_lock);
3404                 return 0;
3405         }
3406         spin_unlock(&fs_info->balance_lock);
3407
3408         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3409                 btrfs_info(fs_info, "force skipping balance");
3410                 return 0;
3411         }
3412
3413         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3414         return PTR_ERR_OR_ZERO(tsk);
3415 }
3416
3417 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3418 {
3419         struct btrfs_balance_control *bctl;
3420         struct btrfs_balance_item *item;
3421         struct btrfs_disk_balance_args disk_bargs;
3422         struct btrfs_path *path;
3423         struct extent_buffer *leaf;
3424         struct btrfs_key key;
3425         int ret;
3426
3427         path = btrfs_alloc_path();
3428         if (!path)
3429                 return -ENOMEM;
3430
3431         key.objectid = BTRFS_BALANCE_OBJECTID;
3432         key.type = BTRFS_BALANCE_ITEM_KEY;
3433         key.offset = 0;
3434
3435         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3436         if (ret < 0)
3437                 goto out;
3438         if (ret > 0) { /* ret = -ENOENT; */
3439                 ret = 0;
3440                 goto out;
3441         }
3442
3443         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3444         if (!bctl) {
3445                 ret = -ENOMEM;
3446                 goto out;
3447         }
3448
3449         leaf = path->nodes[0];
3450         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3451
3452         bctl->fs_info = fs_info;
3453         bctl->flags = btrfs_balance_flags(leaf, item);
3454         bctl->flags |= BTRFS_BALANCE_RESUME;
3455
3456         btrfs_balance_data(leaf, item, &disk_bargs);
3457         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3458         btrfs_balance_meta(leaf, item, &disk_bargs);
3459         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3460         btrfs_balance_sys(leaf, item, &disk_bargs);
3461         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3462
3463         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3464
3465         mutex_lock(&fs_info->volume_mutex);
3466         mutex_lock(&fs_info->balance_mutex);
3467
3468         set_balance_control(bctl);
3469
3470         mutex_unlock(&fs_info->balance_mutex);
3471         mutex_unlock(&fs_info->volume_mutex);
3472 out:
3473         btrfs_free_path(path);
3474         return ret;
3475 }
3476
3477 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3478 {
3479         int ret = 0;
3480
3481         mutex_lock(&fs_info->balance_mutex);
3482         if (!fs_info->balance_ctl) {
3483                 mutex_unlock(&fs_info->balance_mutex);
3484                 return -ENOTCONN;
3485         }
3486
3487         if (atomic_read(&fs_info->balance_running)) {
3488                 atomic_inc(&fs_info->balance_pause_req);
3489                 mutex_unlock(&fs_info->balance_mutex);
3490
3491                 wait_event(fs_info->balance_wait_q,
3492                            atomic_read(&fs_info->balance_running) == 0);
3493
3494                 mutex_lock(&fs_info->balance_mutex);
3495                 /* we are good with balance_ctl ripped off from under us */
3496                 BUG_ON(atomic_read(&fs_info->balance_running));
3497                 atomic_dec(&fs_info->balance_pause_req);
3498         } else {
3499                 ret = -ENOTCONN;
3500         }
3501
3502         mutex_unlock(&fs_info->balance_mutex);
3503         return ret;
3504 }
3505
3506 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3507 {
3508         if (fs_info->sb->s_flags & MS_RDONLY)
3509                 return -EROFS;
3510
3511         mutex_lock(&fs_info->balance_mutex);
3512         if (!fs_info->balance_ctl) {
3513                 mutex_unlock(&fs_info->balance_mutex);
3514                 return -ENOTCONN;
3515         }
3516
3517         atomic_inc(&fs_info->balance_cancel_req);
3518         /*
3519          * if we are running just wait and return, balance item is
3520          * deleted in btrfs_balance in this case
3521          */
3522         if (atomic_read(&fs_info->balance_running)) {
3523                 mutex_unlock(&fs_info->balance_mutex);
3524                 wait_event(fs_info->balance_wait_q,
3525                            atomic_read(&fs_info->balance_running) == 0);
3526                 mutex_lock(&fs_info->balance_mutex);
3527         } else {
3528                 /* __cancel_balance needs volume_mutex */
3529                 mutex_unlock(&fs_info->balance_mutex);
3530                 mutex_lock(&fs_info->volume_mutex);
3531                 mutex_lock(&fs_info->balance_mutex);
3532
3533                 if (fs_info->balance_ctl)
3534                         __cancel_balance(fs_info);
3535
3536                 mutex_unlock(&fs_info->volume_mutex);
3537         }
3538
3539         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3540         atomic_dec(&fs_info->balance_cancel_req);
3541         mutex_unlock(&fs_info->balance_mutex);
3542         return 0;
3543 }
3544
3545 static int btrfs_uuid_scan_kthread(void *data)
3546 {
3547         struct btrfs_fs_info *fs_info = data;
3548         struct btrfs_root *root = fs_info->tree_root;
3549         struct btrfs_key key;
3550         struct btrfs_key max_key;
3551         struct btrfs_path *path = NULL;
3552         int ret = 0;
3553         struct extent_buffer *eb;
3554         int slot;
3555         struct btrfs_root_item root_item;
3556         u32 item_size;
3557         struct btrfs_trans_handle *trans = NULL;
3558
3559         path = btrfs_alloc_path();
3560         if (!path) {
3561                 ret = -ENOMEM;
3562                 goto out;
3563         }
3564
3565         key.objectid = 0;
3566         key.type = BTRFS_ROOT_ITEM_KEY;
3567         key.offset = 0;
3568
3569         max_key.objectid = (u64)-1;
3570         max_key.type = BTRFS_ROOT_ITEM_KEY;
3571         max_key.offset = (u64)-1;
3572
3573         path->keep_locks = 1;
3574
3575         while (1) {
3576                 ret = btrfs_search_forward(root, &key, path, 0);
3577                 if (ret) {
3578                         if (ret > 0)
3579                                 ret = 0;
3580                         break;
3581                 }
3582
3583                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3584                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3585                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3586                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
3587                         goto skip;
3588
3589                 eb = path->nodes[0];
3590                 slot = path->slots[0];
3591                 item_size = btrfs_item_size_nr(eb, slot);
3592                 if (item_size < sizeof(root_item))
3593                         goto skip;
3594
3595                 read_extent_buffer(eb, &root_item,
3596                                    btrfs_item_ptr_offset(eb, slot),
3597                                    (int)sizeof(root_item));
3598                 if (btrfs_root_refs(&root_item) == 0)
3599                         goto skip;
3600
3601                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3602                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
3603                         if (trans)
3604                                 goto update_tree;
3605
3606                         btrfs_release_path(path);
3607                         /*
3608                          * 1 - subvol uuid item
3609                          * 1 - received_subvol uuid item
3610                          */
3611                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3612                         if (IS_ERR(trans)) {
3613                                 ret = PTR_ERR(trans);
3614                                 break;
3615                         }
3616                         continue;
3617                 } else {
3618                         goto skip;
3619                 }
3620 update_tree:
3621                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3622                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3623                                                   root_item.uuid,
3624                                                   BTRFS_UUID_KEY_SUBVOL,
3625                                                   key.objectid);
3626                         if (ret < 0) {
3627                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3628                                         ret);
3629                                 break;
3630                         }
3631                 }
3632
3633                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3634                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3635                                                   root_item.received_uuid,
3636                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3637                                                   key.objectid);
3638                         if (ret < 0) {
3639                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3640                                         ret);
3641                                 break;
3642                         }
3643                 }
3644
3645 skip:
3646                 if (trans) {
3647                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3648                         trans = NULL;
3649                         if (ret)
3650                                 break;
3651                 }
3652
3653                 btrfs_release_path(path);
3654                 if (key.offset < (u64)-1) {
3655                         key.offset++;
3656                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3657                         key.offset = 0;
3658                         key.type = BTRFS_ROOT_ITEM_KEY;
3659                 } else if (key.objectid < (u64)-1) {
3660                         key.offset = 0;
3661                         key.type = BTRFS_ROOT_ITEM_KEY;
3662                         key.objectid++;
3663                 } else {
3664                         break;
3665                 }
3666                 cond_resched();
3667         }
3668
3669 out:
3670         btrfs_free_path(path);
3671         if (trans && !IS_ERR(trans))
3672                 btrfs_end_transaction(trans, fs_info->uuid_root);
3673         if (ret)
3674                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3675         else
3676                 fs_info->update_uuid_tree_gen = 1;
3677         up(&fs_info->uuid_tree_rescan_sem);
3678         return 0;
3679 }
3680
3681 /*
3682  * Callback for btrfs_uuid_tree_iterate().
3683  * returns:
3684  * 0    check succeeded, the entry is not outdated.
3685  * < 0  if an error occured.
3686  * > 0  if the check failed, which means the caller shall remove the entry.
3687  */
3688 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3689                                        u8 *uuid, u8 type, u64 subid)
3690 {
3691         struct btrfs_key key;
3692         int ret = 0;
3693         struct btrfs_root *subvol_root;
3694
3695         if (type != BTRFS_UUID_KEY_SUBVOL &&
3696             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3697                 goto out;
3698
3699         key.objectid = subid;
3700         key.type = BTRFS_ROOT_ITEM_KEY;
3701         key.offset = (u64)-1;
3702         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3703         if (IS_ERR(subvol_root)) {
3704                 ret = PTR_ERR(subvol_root);
3705                 if (ret == -ENOENT)
3706                         ret = 1;
3707                 goto out;
3708         }
3709
3710         switch (type) {
3711         case BTRFS_UUID_KEY_SUBVOL:
3712                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3713                         ret = 1;
3714                 break;
3715         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3716                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3717                            BTRFS_UUID_SIZE))
3718                         ret = 1;
3719                 break;
3720         }
3721
3722 out:
3723         return ret;
3724 }
3725
3726 static int btrfs_uuid_rescan_kthread(void *data)
3727 {
3728         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3729         int ret;
3730
3731         /*
3732          * 1st step is to iterate through the existing UUID tree and
3733          * to delete all entries that contain outdated data.
3734          * 2nd step is to add all missing entries to the UUID tree.
3735          */
3736         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3737         if (ret < 0) {
3738                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3739                 up(&fs_info->uuid_tree_rescan_sem);
3740                 return ret;
3741         }
3742         return btrfs_uuid_scan_kthread(data);
3743 }
3744
3745 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3746 {
3747         struct btrfs_trans_handle *trans;
3748         struct btrfs_root *tree_root = fs_info->tree_root;
3749         struct btrfs_root *uuid_root;
3750         struct task_struct *task;
3751         int ret;
3752
3753         /*
3754          * 1 - root node
3755          * 1 - root item
3756          */
3757         trans = btrfs_start_transaction(tree_root, 2);
3758         if (IS_ERR(trans))
3759                 return PTR_ERR(trans);
3760
3761         uuid_root = btrfs_create_tree(trans, fs_info,
3762                                       BTRFS_UUID_TREE_OBJECTID);
3763         if (IS_ERR(uuid_root)) {
3764                 btrfs_abort_transaction(trans, tree_root,
3765                                         PTR_ERR(uuid_root));
3766                 return PTR_ERR(uuid_root);
3767         }
3768
3769         fs_info->uuid_root = uuid_root;
3770
3771         ret = btrfs_commit_transaction(trans, tree_root);
3772         if (ret)
3773                 return ret;
3774
3775         down(&fs_info->uuid_tree_rescan_sem);
3776         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3777         if (IS_ERR(task)) {
3778                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3779                 btrfs_warn(fs_info, "failed to start uuid_scan task");
3780                 up(&fs_info->uuid_tree_rescan_sem);
3781                 return PTR_ERR(task);
3782         }
3783
3784         return 0;
3785 }
3786
3787 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3788 {
3789         struct task_struct *task;
3790
3791         down(&fs_info->uuid_tree_rescan_sem);
3792         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3793         if (IS_ERR(task)) {
3794                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3795                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3796                 up(&fs_info->uuid_tree_rescan_sem);
3797                 return PTR_ERR(task);
3798         }
3799
3800         return 0;
3801 }
3802
3803 /*
3804  * shrinking a device means finding all of the device extents past
3805  * the new size, and then following the back refs to the chunks.
3806  * The chunk relocation code actually frees the device extent
3807  */
3808 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3809 {
3810         struct btrfs_trans_handle *trans;
3811         struct btrfs_root *root = device->dev_root;
3812         struct btrfs_dev_extent *dev_extent = NULL;
3813         struct btrfs_path *path;
3814         u64 length;
3815         u64 chunk_tree;
3816         u64 chunk_objectid;
3817         u64 chunk_offset;
3818         int ret;
3819         int slot;
3820         int failed = 0;
3821         bool retried = false;
3822         struct extent_buffer *l;
3823         struct btrfs_key key;
3824         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3825         u64 old_total = btrfs_super_total_bytes(super_copy);
3826         u64 old_size = device->total_bytes;
3827         u64 diff = device->total_bytes - new_size;
3828
3829         if (device->is_tgtdev_for_dev_replace)
3830                 return -EINVAL;
3831
3832         path = btrfs_alloc_path();
3833         if (!path)
3834                 return -ENOMEM;
3835
3836         path->reada = 2;
3837
3838         lock_chunks(root);
3839
3840         device->total_bytes = new_size;
3841         if (device->writeable) {
3842                 device->fs_devices->total_rw_bytes -= diff;
3843                 spin_lock(&root->fs_info->free_chunk_lock);
3844                 root->fs_info->free_chunk_space -= diff;
3845                 spin_unlock(&root->fs_info->free_chunk_lock);
3846         }
3847         unlock_chunks(root);
3848
3849 again:
3850         key.objectid = device->devid;
3851         key.offset = (u64)-1;
3852         key.type = BTRFS_DEV_EXTENT_KEY;
3853
3854         do {
3855                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3856                 if (ret < 0)
3857                         goto done;
3858
3859                 ret = btrfs_previous_item(root, path, 0, key.type);
3860                 if (ret < 0)
3861                         goto done;
3862                 if (ret) {
3863                         ret = 0;
3864                         btrfs_release_path(path);
3865                         break;
3866                 }
3867
3868                 l = path->nodes[0];
3869                 slot = path->slots[0];
3870                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3871
3872                 if (key.objectid != device->devid) {
3873                         btrfs_release_path(path);
3874                         break;
3875                 }
3876
3877                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3878                 length = btrfs_dev_extent_length(l, dev_extent);
3879
3880                 if (key.offset + length <= new_size) {
3881                         btrfs_release_path(path);
3882                         break;
3883                 }
3884
3885                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3886                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3887                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3888                 btrfs_release_path(path);
3889
3890                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3891                                            chunk_offset);
3892                 if (ret && ret != -ENOSPC)
3893                         goto done;
3894                 if (ret == -ENOSPC)
3895                         failed++;
3896         } while (key.offset-- > 0);
3897
3898         if (failed && !retried) {
3899                 failed = 0;
3900                 retried = true;
3901                 goto again;
3902         } else if (failed && retried) {
3903                 ret = -ENOSPC;
3904                 lock_chunks(root);
3905
3906                 device->total_bytes = old_size;
3907                 if (device->writeable)
3908                         device->fs_devices->total_rw_bytes += diff;
3909                 spin_lock(&root->fs_info->free_chunk_lock);
3910                 root->fs_info->free_chunk_space += diff;
3911                 spin_unlock(&root->fs_info->free_chunk_lock);
3912                 unlock_chunks(root);
3913                 goto done;
3914         }
3915
3916         /* Shrinking succeeded, else we would be at "done". */
3917         trans = btrfs_start_transaction(root, 0);
3918         if (IS_ERR(trans)) {
3919                 ret = PTR_ERR(trans);
3920                 goto done;
3921         }
3922
3923         lock_chunks(root);
3924
3925         device->disk_total_bytes = new_size;
3926         /* Now btrfs_update_device() will change the on-disk size. */
3927         ret = btrfs_update_device(trans, device);
3928         if (ret) {
3929                 unlock_chunks(root);
3930                 btrfs_end_transaction(trans, root);
3931                 goto done;
3932         }
3933         WARN_ON(diff > old_total);
3934         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3935         unlock_chunks(root);
3936         btrfs_end_transaction(trans, root);
3937 done:
3938         btrfs_free_path(path);
3939         return ret;
3940 }
3941
3942 static int btrfs_add_system_chunk(struct btrfs_root *root,
3943                            struct btrfs_key *key,
3944                            struct btrfs_chunk *chunk, int item_size)
3945 {
3946         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3947         struct btrfs_disk_key disk_key;
3948         u32 array_size;
3949         u8 *ptr;
3950
3951         array_size = btrfs_super_sys_array_size(super_copy);
3952         if (array_size + item_size + sizeof(disk_key)
3953                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3954                 return -EFBIG;
3955
3956         ptr = super_copy->sys_chunk_array + array_size;
3957         btrfs_cpu_key_to_disk(&disk_key, key);
3958         memcpy(ptr, &disk_key, sizeof(disk_key));
3959         ptr += sizeof(disk_key);
3960         memcpy(ptr, chunk, item_size);
3961         item_size += sizeof(disk_key);
3962         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3963         return 0;
3964 }
3965
3966 /*
3967  * sort the devices in descending order by max_avail, total_avail
3968  */
3969 static int btrfs_cmp_device_info(const void *a, const void *b)
3970 {
3971         const struct btrfs_device_info *di_a = a;
3972         const struct btrfs_device_info *di_b = b;
3973
3974         if (di_a->max_avail > di_b->max_avail)
3975                 return -1;
3976         if (di_a->max_avail < di_b->max_avail)
3977                 return 1;
3978         if (di_a->total_avail > di_b->total_avail)
3979                 return -1;
3980         if (di_a->total_avail < di_b->total_avail)
3981                 return 1;
3982         return 0;
3983 }
3984
3985 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3986         [BTRFS_RAID_RAID10] = {
3987                 .sub_stripes    = 2,
3988                 .dev_stripes    = 1,
3989                 .devs_max       = 0,    /* 0 == as many as possible */
3990                 .devs_min       = 4,
3991                 .devs_increment = 2,
3992                 .ncopies        = 2,
3993         },
3994         [BTRFS_RAID_RAID1] = {
3995                 .sub_stripes    = 1,
3996                 .dev_stripes    = 1,
3997                 .devs_max       = 2,
3998                 .devs_min       = 2,
3999                 .devs_increment = 2,
4000                 .ncopies        = 2,
4001         },
4002         [BTRFS_RAID_DUP] = {
4003                 .sub_stripes    = 1,
4004                 .dev_stripes    = 2,
4005                 .devs_max       = 1,
4006                 .devs_min       = 1,
4007                 .devs_increment = 1,
4008                 .ncopies        = 2,
4009         },
4010         [BTRFS_RAID_RAID0] = {
4011                 .sub_stripes    = 1,
4012                 .dev_stripes    = 1,
4013                 .devs_max       = 0,
4014                 .devs_min       = 2,
4015                 .devs_increment = 1,
4016                 .ncopies        = 1,
4017         },
4018         [BTRFS_RAID_SINGLE] = {
4019                 .sub_stripes    = 1,
4020                 .dev_stripes    = 1,
4021                 .devs_max       = 1,
4022                 .devs_min       = 1,
4023                 .devs_increment = 1,
4024                 .ncopies        = 1,
4025         },
4026         [BTRFS_RAID_RAID5] = {
4027                 .sub_stripes    = 1,
4028                 .dev_stripes    = 1,
4029                 .devs_max       = 0,
4030                 .devs_min       = 2,
4031                 .devs_increment = 1,
4032                 .ncopies        = 2,
4033         },
4034         [BTRFS_RAID_RAID6] = {
4035                 .sub_stripes    = 1,
4036                 .dev_stripes    = 1,
4037                 .devs_max       = 0,
4038                 .devs_min       = 3,
4039                 .devs_increment = 1,
4040                 .ncopies        = 3,
4041         },
4042 };
4043
4044 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4045 {
4046         /* TODO allow them to set a preferred stripe size */
4047         return 64 * 1024;
4048 }
4049
4050 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4051 {
4052         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
4053                 return;
4054
4055         btrfs_set_fs_incompat(info, RAID56);
4056 }
4057
4058 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4059                         - sizeof(struct btrfs_item)             \
4060                         - sizeof(struct btrfs_chunk))           \
4061                         / sizeof(struct btrfs_stripe) + 1)
4062
4063 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4064                                 - 2 * sizeof(struct btrfs_disk_key)     \
4065                                 - 2 * sizeof(struct btrfs_chunk))       \
4066                                 / sizeof(struct btrfs_stripe) + 1)
4067
4068 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4069                                struct btrfs_root *extent_root, u64 start,
4070                                u64 type)
4071 {
4072         struct btrfs_fs_info *info = extent_root->fs_info;
4073         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4074         struct list_head *cur;
4075         struct map_lookup *map = NULL;
4076         struct extent_map_tree *em_tree;
4077         struct extent_map *em;
4078         struct btrfs_device_info *devices_info = NULL;
4079         u64 total_avail;
4080         int num_stripes;        /* total number of stripes to allocate */
4081         int data_stripes;       /* number of stripes that count for
4082                                    block group size */
4083         int sub_stripes;        /* sub_stripes info for map */
4084         int dev_stripes;        /* stripes per dev */
4085         int devs_max;           /* max devs to use */
4086         int devs_min;           /* min devs needed */
4087         int devs_increment;     /* ndevs has to be a multiple of this */
4088         int ncopies;            /* how many copies to data has */
4089         int ret;
4090         u64 max_stripe_size;
4091         u64 max_chunk_size;
4092         u64 stripe_size;
4093         u64 num_bytes;
4094         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4095         int ndevs;
4096         int i;
4097         int j;
4098         int index;
4099
4100         BUG_ON(!alloc_profile_is_valid(type, 0));
4101
4102         if (list_empty(&fs_devices->alloc_list))
4103                 return -ENOSPC;
4104
4105         index = __get_raid_index(type);
4106
4107         sub_stripes = btrfs_raid_array[index].sub_stripes;
4108         dev_stripes = btrfs_raid_array[index].dev_stripes;
4109         devs_max = btrfs_raid_array[index].devs_max;
4110         devs_min = btrfs_raid_array[index].devs_min;
4111         devs_increment = btrfs_raid_array[index].devs_increment;
4112         ncopies = btrfs_raid_array[index].ncopies;
4113
4114         if (type & BTRFS_BLOCK_GROUP_DATA) {
4115                 max_stripe_size = 1024 * 1024 * 1024;
4116                 max_chunk_size = 10 * max_stripe_size;
4117                 if (!devs_max)
4118                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4119         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4120                 /* for larger filesystems, use larger metadata chunks */
4121                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4122                         max_stripe_size = 1024 * 1024 * 1024;
4123                 else
4124                         max_stripe_size = 256 * 1024 * 1024;
4125                 max_chunk_size = max_stripe_size;
4126                 if (!devs_max)
4127                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4128         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4129                 max_stripe_size = 32 * 1024 * 1024;
4130                 max_chunk_size = 2 * max_stripe_size;
4131                 if (!devs_max)
4132                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4133         } else {
4134                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4135                        type);
4136                 BUG_ON(1);
4137         }
4138
4139         /* we don't want a chunk larger than 10% of writeable space */
4140         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4141                              max_chunk_size);
4142
4143         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4144                                GFP_NOFS);
4145         if (!devices_info)
4146                 return -ENOMEM;
4147
4148         cur = fs_devices->alloc_list.next;
4149
4150         /*
4151          * in the first pass through the devices list, we gather information
4152          * about the available holes on each device.
4153          */
4154         ndevs = 0;
4155         while (cur != &fs_devices->alloc_list) {
4156                 struct btrfs_device *device;
4157                 u64 max_avail;
4158                 u64 dev_offset;
4159
4160                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4161
4162                 cur = cur->next;
4163
4164                 if (!device->writeable) {
4165                         WARN(1, KERN_ERR
4166                                "BTRFS: read-only device in alloc_list\n");
4167                         continue;
4168                 }
4169
4170                 if (!device->in_fs_metadata ||
4171                     device->is_tgtdev_for_dev_replace)
4172                         continue;
4173
4174                 if (device->total_bytes > device->bytes_used)
4175                         total_avail = device->total_bytes - device->bytes_used;
4176                 else
4177                         total_avail = 0;
4178
4179                 /* If there is no space on this device, skip it. */
4180                 if (total_avail == 0)
4181                         continue;
4182
4183                 ret = find_free_dev_extent(trans, device,
4184                                            max_stripe_size * dev_stripes,
4185                                            &dev_offset, &max_avail);
4186                 if (ret && ret != -ENOSPC)
4187                         goto error;
4188
4189                 if (ret == 0)
4190                         max_avail = max_stripe_size * dev_stripes;
4191
4192                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4193                         continue;
4194
4195                 if (ndevs == fs_devices->rw_devices) {
4196                         WARN(1, "%s: found more than %llu devices\n",
4197                              __func__, fs_devices->rw_devices);
4198                         break;
4199                 }
4200                 devices_info[ndevs].dev_offset = dev_offset;
4201                 devices_info[ndevs].max_avail = max_avail;
4202                 devices_info[ndevs].total_avail = total_avail;
4203                 devices_info[ndevs].dev = device;
4204                 ++ndevs;
4205         }
4206
4207         /*
4208          * now sort the devices by hole size / available space
4209          */
4210         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4211              btrfs_cmp_device_info, NULL);
4212
4213         /* round down to number of usable stripes */
4214         ndevs -= ndevs % devs_increment;
4215
4216         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4217                 ret = -ENOSPC;
4218                 goto error;
4219         }
4220
4221         if (devs_max && ndevs > devs_max)
4222                 ndevs = devs_max;
4223         /*
4224          * the primary goal is to maximize the number of stripes, so use as many
4225          * devices as possible, even if the stripes are not maximum sized.
4226          */
4227         stripe_size = devices_info[ndevs-1].max_avail;
4228         num_stripes = ndevs * dev_stripes;
4229
4230         /*
4231          * this will have to be fixed for RAID1 and RAID10 over
4232          * more drives
4233          */
4234         data_stripes = num_stripes / ncopies;
4235
4236         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4237                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4238                                  btrfs_super_stripesize(info->super_copy));
4239                 data_stripes = num_stripes - 1;
4240         }
4241         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4242                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4243                                  btrfs_super_stripesize(info->super_copy));
4244                 data_stripes = num_stripes - 2;
4245         }
4246
4247         /*
4248          * Use the number of data stripes to figure out how big this chunk
4249          * is really going to be in terms of logical address space,
4250          * and compare that answer with the max chunk size
4251          */
4252         if (stripe_size * data_stripes > max_chunk_size) {
4253                 u64 mask = (1ULL << 24) - 1;
4254                 stripe_size = max_chunk_size;
4255                 do_div(stripe_size, data_stripes);
4256
4257                 /* bump the answer up to a 16MB boundary */
4258                 stripe_size = (stripe_size + mask) & ~mask;
4259
4260                 /* but don't go higher than the limits we found
4261                  * while searching for free extents
4262                  */
4263                 if (stripe_size > devices_info[ndevs-1].max_avail)
4264                         stripe_size = devices_info[ndevs-1].max_avail;
4265         }
4266
4267         do_div(stripe_size, dev_stripes);
4268
4269         /* align to BTRFS_STRIPE_LEN */
4270         do_div(stripe_size, raid_stripe_len);
4271         stripe_size *= raid_stripe_len;
4272
4273         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4274         if (!map) {
4275                 ret = -ENOMEM;
4276                 goto error;
4277         }
4278         map->num_stripes = num_stripes;
4279
4280         for (i = 0; i < ndevs; ++i) {
4281                 for (j = 0; j < dev_stripes; ++j) {
4282                         int s = i * dev_stripes + j;
4283                         map->stripes[s].dev = devices_info[i].dev;
4284                         map->stripes[s].physical = devices_info[i].dev_offset +
4285                                                    j * stripe_size;
4286                 }
4287         }
4288         map->sector_size = extent_root->sectorsize;
4289         map->stripe_len = raid_stripe_len;
4290         map->io_align = raid_stripe_len;
4291         map->io_width = raid_stripe_len;
4292         map->type = type;
4293         map->sub_stripes = sub_stripes;
4294
4295         num_bytes = stripe_size * data_stripes;
4296
4297         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4298
4299         em = alloc_extent_map();
4300         if (!em) {
4301                 kfree(map);
4302                 ret = -ENOMEM;
4303                 goto error;
4304         }
4305         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4306         em->bdev = (struct block_device *)map;
4307         em->start = start;
4308         em->len = num_bytes;
4309         em->block_start = 0;
4310         em->block_len = em->len;
4311         em->orig_block_len = stripe_size;
4312
4313         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4314         write_lock(&em_tree->lock);
4315         ret = add_extent_mapping(em_tree, em, 0);
4316         if (!ret) {
4317                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4318                 atomic_inc(&em->refs);
4319         }
4320         write_unlock(&em_tree->lock);
4321         if (ret) {
4322                 free_extent_map(em);
4323                 goto error;
4324         }
4325
4326         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4327                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4328                                      start, num_bytes);
4329         if (ret)
4330                 goto error_del_extent;
4331
4332         free_extent_map(em);
4333         check_raid56_incompat_flag(extent_root->fs_info, type);
4334
4335         kfree(devices_info);
4336         return 0;
4337
4338 error_del_extent:
4339         write_lock(&em_tree->lock);
4340         remove_extent_mapping(em_tree, em);
4341         write_unlock(&em_tree->lock);
4342
4343         /* One for our allocation */
4344         free_extent_map(em);
4345         /* One for the tree reference */
4346         free_extent_map(em);
4347 error:
4348         kfree(devices_info);
4349         return ret;
4350 }
4351
4352 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4353                                 struct btrfs_root *extent_root,
4354                                 u64 chunk_offset, u64 chunk_size)
4355 {
4356         struct btrfs_key key;
4357         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4358         struct btrfs_device *device;
4359         struct btrfs_chunk *chunk;
4360         struct btrfs_stripe *stripe;
4361         struct extent_map_tree *em_tree;
4362         struct extent_map *em;
4363         struct map_lookup *map;
4364         size_t item_size;
4365         u64 dev_offset;
4366         u64 stripe_size;
4367         int i = 0;
4368         int ret;
4369
4370         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4371         read_lock(&em_tree->lock);
4372         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4373         read_unlock(&em_tree->lock);
4374
4375         if (!em) {
4376                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4377                            "%Lu len %Lu", chunk_offset, chunk_size);
4378                 return -EINVAL;
4379         }
4380
4381         if (em->start != chunk_offset || em->len != chunk_size) {
4382                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4383                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4384                           chunk_size, em->start, em->len);
4385                 free_extent_map(em);
4386                 return -EINVAL;
4387         }
4388
4389         map = (struct map_lookup *)em->bdev;
4390         item_size = btrfs_chunk_item_size(map->num_stripes);
4391         stripe_size = em->orig_block_len;
4392
4393         chunk = kzalloc(item_size, GFP_NOFS);
4394         if (!chunk) {
4395                 ret = -ENOMEM;
4396                 goto out;
4397         }
4398
4399         for (i = 0; i < map->num_stripes; i++) {
4400                 device = map->stripes[i].dev;
4401                 dev_offset = map->stripes[i].physical;
4402
4403                 device->bytes_used += stripe_size;
4404                 ret = btrfs_update_device(trans, device);
4405                 if (ret)
4406                         goto out;
4407                 ret = btrfs_alloc_dev_extent(trans, device,
4408                                              chunk_root->root_key.objectid,
4409                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4410                                              chunk_offset, dev_offset,
4411                                              stripe_size);
4412                 if (ret)
4413                         goto out;
4414         }
4415
4416         spin_lock(&extent_root->fs_info->free_chunk_lock);
4417         extent_root->fs_info->free_chunk_space -= (stripe_size *
4418                                                    map->num_stripes);
4419         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4420
4421         stripe = &chunk->stripe;
4422         for (i = 0; i < map->num_stripes; i++) {
4423                 device = map->stripes[i].dev;
4424                 dev_offset = map->stripes[i].physical;
4425
4426                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4427                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4428                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4429                 stripe++;
4430         }
4431
4432         btrfs_set_stack_chunk_length(chunk, chunk_size);
4433         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4434         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4435         btrfs_set_stack_chunk_type(chunk, map->type);
4436         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4437         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4438         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4439         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4440         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4441
4442         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4443         key.type = BTRFS_CHUNK_ITEM_KEY;
4444         key.offset = chunk_offset;
4445
4446         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4447         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4448                 /*
4449                  * TODO: Cleanup of inserted chunk root in case of
4450                  * failure.
4451                  */
4452                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4453                                              item_size);
4454         }
4455
4456 out:
4457         kfree(chunk);
4458         free_extent_map(em);
4459         return ret;
4460 }
4461
4462 /*
4463  * Chunk allocation falls into two parts. The first part does works
4464  * that make the new allocated chunk useable, but not do any operation
4465  * that modifies the chunk tree. The second part does the works that
4466  * require modifying the chunk tree. This division is important for the
4467  * bootstrap process of adding storage to a seed btrfs.
4468  */
4469 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4470                       struct btrfs_root *extent_root, u64 type)
4471 {
4472         u64 chunk_offset;
4473
4474         chunk_offset = find_next_chunk(extent_root->fs_info);
4475         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4476 }
4477
4478 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4479                                          struct btrfs_root *root,
4480                                          struct btrfs_device *device)
4481 {
4482         u64 chunk_offset;
4483         u64 sys_chunk_offset;
4484         u64 alloc_profile;
4485         struct btrfs_fs_info *fs_info = root->fs_info;
4486         struct btrfs_root *extent_root = fs_info->extent_root;
4487         int ret;
4488
4489         chunk_offset = find_next_chunk(fs_info);
4490         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4491         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4492                                   alloc_profile);
4493         if (ret)
4494                 return ret;
4495
4496         sys_chunk_offset = find_next_chunk(root->fs_info);
4497         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4498         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4499                                   alloc_profile);
4500         if (ret) {
4501                 btrfs_abort_transaction(trans, root, ret);
4502                 goto out;
4503         }
4504
4505         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4506         if (ret)
4507                 btrfs_abort_transaction(trans, root, ret);
4508 out:
4509         return ret;
4510 }
4511
4512 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4513 {
4514         struct extent_map *em;
4515         struct map_lookup *map;
4516         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4517         int readonly = 0;
4518         int i;
4519
4520         read_lock(&map_tree->map_tree.lock);
4521         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4522         read_unlock(&map_tree->map_tree.lock);
4523         if (!em)
4524                 return 1;
4525
4526         if (btrfs_test_opt(root, DEGRADED)) {
4527                 free_extent_map(em);
4528                 return 0;
4529         }
4530
4531         map = (struct map_lookup *)em->bdev;
4532         for (i = 0; i < map->num_stripes; i++) {
4533                 if (!map->stripes[i].dev->writeable) {
4534                         readonly = 1;
4535                         break;
4536                 }
4537         }
4538         free_extent_map(em);
4539         return readonly;
4540 }
4541
4542 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4543 {
4544         extent_map_tree_init(&tree->map_tree);
4545 }
4546
4547 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4548 {
4549         struct extent_map *em;
4550
4551         while (1) {
4552                 write_lock(&tree->map_tree.lock);
4553                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4554                 if (em)
4555                         remove_extent_mapping(&tree->map_tree, em);
4556                 write_unlock(&tree->map_tree.lock);
4557                 if (!em)
4558                         break;
4559                 /* once for us */
4560                 free_extent_map(em);
4561                 /* once for the tree */
4562                 free_extent_map(em);
4563         }
4564 }
4565
4566 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4567 {
4568         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4569         struct extent_map *em;
4570         struct map_lookup *map;
4571         struct extent_map_tree *em_tree = &map_tree->map_tree;
4572         int ret;
4573
4574         read_lock(&em_tree->lock);
4575         em = lookup_extent_mapping(em_tree, logical, len);
4576         read_unlock(&em_tree->lock);
4577
4578         /*
4579          * We could return errors for these cases, but that could get ugly and
4580          * we'd probably do the same thing which is just not do anything else
4581          * and exit, so return 1 so the callers don't try to use other copies.
4582          */
4583         if (!em) {
4584                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4585                             logical+len);
4586                 return 1;
4587         }
4588
4589         if (em->start > logical || em->start + em->len < logical) {
4590                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4591                             "%Lu-%Lu", logical, logical+len, em->start,
4592                             em->start + em->len);
4593                 free_extent_map(em);
4594                 return 1;
4595         }
4596
4597         map = (struct map_lookup *)em->bdev;
4598         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4599                 ret = map->num_stripes;
4600         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4601                 ret = map->sub_stripes;
4602         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4603                 ret = 2;
4604         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4605                 ret = 3;
4606         else
4607                 ret = 1;
4608         free_extent_map(em);
4609
4610         btrfs_dev_replace_lock(&fs_info->dev_replace);
4611         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4612                 ret++;
4613         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4614
4615         return ret;
4616 }
4617
4618 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4619                                     struct btrfs_mapping_tree *map_tree,
4620                                     u64 logical)
4621 {
4622         struct extent_map *em;
4623         struct map_lookup *map;
4624         struct extent_map_tree *em_tree = &map_tree->map_tree;
4625         unsigned long len = root->sectorsize;
4626
4627         read_lock(&em_tree->lock);
4628         em = lookup_extent_mapping(em_tree, logical, len);
4629         read_unlock(&em_tree->lock);
4630         BUG_ON(!em);
4631
4632         BUG_ON(em->start > logical || em->start + em->len < logical);
4633         map = (struct map_lookup *)em->bdev;
4634         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4635                          BTRFS_BLOCK_GROUP_RAID6)) {
4636                 len = map->stripe_len * nr_data_stripes(map);
4637         }
4638         free_extent_map(em);
4639         return len;
4640 }
4641
4642 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4643                            u64 logical, u64 len, int mirror_num)
4644 {
4645         struct extent_map *em;
4646         struct map_lookup *map;
4647         struct extent_map_tree *em_tree = &map_tree->map_tree;
4648         int ret = 0;
4649
4650         read_lock(&em_tree->lock);
4651         em = lookup_extent_mapping(em_tree, logical, len);
4652         read_unlock(&em_tree->lock);
4653         BUG_ON(!em);
4654
4655         BUG_ON(em->start > logical || em->start + em->len < logical);
4656         map = (struct map_lookup *)em->bdev;
4657         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4658                          BTRFS_BLOCK_GROUP_RAID6))
4659                 ret = 1;
4660         free_extent_map(em);
4661         return ret;
4662 }
4663
4664 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4665                             struct map_lookup *map, int first, int num,
4666                             int optimal, int dev_replace_is_ongoing)
4667 {
4668         int i;
4669         int tolerance;
4670         struct btrfs_device *srcdev;
4671
4672         if (dev_replace_is_ongoing &&
4673             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4674              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4675                 srcdev = fs_info->dev_replace.srcdev;
4676         else
4677                 srcdev = NULL;
4678
4679         /*
4680          * try to avoid the drive that is the source drive for a
4681          * dev-replace procedure, only choose it if no other non-missing
4682          * mirror is available
4683          */
4684         for (tolerance = 0; tolerance < 2; tolerance++) {
4685                 if (map->stripes[optimal].dev->bdev &&
4686                     (tolerance || map->stripes[optimal].dev != srcdev))
4687                         return optimal;
4688                 for (i = first; i < first + num; i++) {
4689                         if (map->stripes[i].dev->bdev &&
4690                             (tolerance || map->stripes[i].dev != srcdev))
4691                                 return i;
4692                 }
4693         }
4694
4695         /* we couldn't find one that doesn't fail.  Just return something
4696          * and the io error handling code will clean up eventually
4697          */
4698         return optimal;
4699 }
4700
4701 static inline int parity_smaller(u64 a, u64 b)
4702 {
4703         return a > b;
4704 }
4705
4706 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4707 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4708 {
4709         struct btrfs_bio_stripe s;
4710         int i;
4711         u64 l;
4712         int again = 1;
4713
4714         while (again) {
4715                 again = 0;
4716                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4717                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4718                                 s = bbio->stripes[i];
4719                                 l = raid_map[i];
4720                                 bbio->stripes[i] = bbio->stripes[i+1];
4721                                 raid_map[i] = raid_map[i+1];
4722                                 bbio->stripes[i+1] = s;
4723                                 raid_map[i+1] = l;
4724                                 again = 1;
4725                         }
4726                 }
4727         }
4728 }
4729
4730 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4731                              u64 logical, u64 *length,
4732                              struct btrfs_bio **bbio_ret,
4733                              int mirror_num, u64 **raid_map_ret)
4734 {
4735         struct extent_map *em;
4736         struct map_lookup *map;
4737         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4738         struct extent_map_tree *em_tree = &map_tree->map_tree;
4739         u64 offset;
4740         u64 stripe_offset;
4741         u64 stripe_end_offset;
4742         u64 stripe_nr;
4743         u64 stripe_nr_orig;
4744         u64 stripe_nr_end;
4745         u64 stripe_len;
4746         u64 *raid_map = NULL;
4747         int stripe_index;
4748         int i;
4749         int ret = 0;
4750         int num_stripes;
4751         int max_errors = 0;
4752         struct btrfs_bio *bbio = NULL;
4753         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4754         int dev_replace_is_ongoing = 0;
4755         int num_alloc_stripes;
4756         int patch_the_first_stripe_for_dev_replace = 0;
4757         u64 physical_to_patch_in_first_stripe = 0;
4758         u64 raid56_full_stripe_start = (u64)-1;
4759
4760         read_lock(&em_tree->lock);
4761         em = lookup_extent_mapping(em_tree, logical, *length);
4762         read_unlock(&em_tree->lock);
4763
4764         if (!em) {
4765                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4766                         logical, *length);
4767                 return -EINVAL;
4768         }
4769
4770         if (em->start > logical || em->start + em->len < logical) {
4771                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4772                            "found %Lu-%Lu", logical, em->start,
4773                            em->start + em->len);
4774                 free_extent_map(em);
4775                 return -EINVAL;
4776         }
4777
4778         map = (struct map_lookup *)em->bdev;
4779         offset = logical - em->start;
4780
4781         stripe_len = map->stripe_len;
4782         stripe_nr = offset;
4783         /*
4784          * stripe_nr counts the total number of stripes we have to stride
4785          * to get to this block
4786          */
4787         do_div(stripe_nr, stripe_len);
4788
4789         stripe_offset = stripe_nr * stripe_len;
4790         BUG_ON(offset < stripe_offset);
4791
4792         /* stripe_offset is the offset of this block in its stripe*/
4793         stripe_offset = offset - stripe_offset;
4794
4795         /* if we're here for raid56, we need to know the stripe aligned start */
4796         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4797                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4798                 raid56_full_stripe_start = offset;
4799
4800                 /* allow a write of a full stripe, but make sure we don't
4801                  * allow straddling of stripes
4802                  */
4803                 do_div(raid56_full_stripe_start, full_stripe_len);
4804                 raid56_full_stripe_start *= full_stripe_len;
4805         }
4806
4807         if (rw & REQ_DISCARD) {
4808                 /* we don't discard raid56 yet */
4809                 if (map->type &
4810                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4811                         ret = -EOPNOTSUPP;
4812                         goto out;
4813                 }
4814                 *length = min_t(u64, em->len - offset, *length);
4815         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4816                 u64 max_len;
4817                 /* For writes to RAID[56], allow a full stripeset across all disks.
4818                    For other RAID types and for RAID[56] reads, just allow a single
4819                    stripe (on a single disk). */
4820                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4821                     (rw & REQ_WRITE)) {
4822                         max_len = stripe_len * nr_data_stripes(map) -
4823                                 (offset - raid56_full_stripe_start);
4824                 } else {
4825                         /* we limit the length of each bio to what fits in a stripe */
4826                         max_len = stripe_len - stripe_offset;
4827                 }
4828                 *length = min_t(u64, em->len - offset, max_len);
4829         } else {
4830                 *length = em->len - offset;
4831         }
4832
4833         /* This is for when we're called from btrfs_merge_bio_hook() and all
4834            it cares about is the length */
4835         if (!bbio_ret)
4836                 goto out;
4837
4838         btrfs_dev_replace_lock(dev_replace);
4839         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4840         if (!dev_replace_is_ongoing)
4841                 btrfs_dev_replace_unlock(dev_replace);
4842
4843         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4844             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4845             dev_replace->tgtdev != NULL) {
4846                 /*
4847                  * in dev-replace case, for repair case (that's the only
4848                  * case where the mirror is selected explicitly when
4849                  * calling btrfs_map_block), blocks left of the left cursor
4850                  * can also be read from the target drive.
4851                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4852                  * the last one to the array of stripes. For READ, it also
4853                  * needs to be supported using the same mirror number.
4854                  * If the requested block is not left of the left cursor,
4855                  * EIO is returned. This can happen because btrfs_num_copies()
4856                  * returns one more in the dev-replace case.
4857                  */
4858                 u64 tmp_length = *length;
4859                 struct btrfs_bio *tmp_bbio = NULL;
4860                 int tmp_num_stripes;
4861                 u64 srcdev_devid = dev_replace->srcdev->devid;
4862                 int index_srcdev = 0;
4863                 int found = 0;
4864                 u64 physical_of_found = 0;
4865
4866                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4867                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4868                 if (ret) {
4869                         WARN_ON(tmp_bbio != NULL);
4870                         goto out;
4871                 }
4872
4873                 tmp_num_stripes = tmp_bbio->num_stripes;
4874                 if (mirror_num > tmp_num_stripes) {
4875                         /*
4876                          * REQ_GET_READ_MIRRORS does not contain this
4877                          * mirror, that means that the requested area
4878                          * is not left of the left cursor
4879                          */
4880                         ret = -EIO;
4881                         kfree(tmp_bbio);
4882                         goto out;
4883                 }
4884
4885                 /*
4886                  * process the rest of the function using the mirror_num
4887                  * of the source drive. Therefore look it up first.
4888                  * At the end, patch the device pointer to the one of the
4889                  * target drive.
4890                  */
4891                 for (i = 0; i < tmp_num_stripes; i++) {
4892                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4893                                 /*
4894                                  * In case of DUP, in order to keep it
4895                                  * simple, only add the mirror with the
4896                                  * lowest physical address
4897                                  */
4898                                 if (found &&
4899                                     physical_of_found <=
4900                                      tmp_bbio->stripes[i].physical)
4901                                         continue;
4902                                 index_srcdev = i;
4903                                 found = 1;
4904                                 physical_of_found =
4905                                         tmp_bbio->stripes[i].physical;
4906                         }
4907                 }
4908
4909                 if (found) {
4910                         mirror_num = index_srcdev + 1;
4911                         patch_the_first_stripe_for_dev_replace = 1;
4912                         physical_to_patch_in_first_stripe = physical_of_found;
4913                 } else {
4914                         WARN_ON(1);
4915                         ret = -EIO;
4916                         kfree(tmp_bbio);
4917                         goto out;
4918                 }
4919
4920                 kfree(tmp_bbio);
4921         } else if (mirror_num > map->num_stripes) {
4922                 mirror_num = 0;
4923         }
4924
4925         num_stripes = 1;
4926         stripe_index = 0;
4927         stripe_nr_orig = stripe_nr;
4928         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4929         do_div(stripe_nr_end, map->stripe_len);
4930         stripe_end_offset = stripe_nr_end * map->stripe_len -
4931                             (offset + *length);
4932
4933         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4934                 if (rw & REQ_DISCARD)
4935                         num_stripes = min_t(u64, map->num_stripes,
4936                                             stripe_nr_end - stripe_nr_orig);
4937                 stripe_index = do_div(stripe_nr, map->num_stripes);
4938         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4939                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4940                         num_stripes = map->num_stripes;
4941                 else if (mirror_num)
4942                         stripe_index = mirror_num - 1;
4943                 else {
4944                         stripe_index = find_live_mirror(fs_info, map, 0,
4945                                             map->num_stripes,
4946                                             current->pid % map->num_stripes,
4947                                             dev_replace_is_ongoing);
4948                         mirror_num = stripe_index + 1;
4949                 }
4950
4951         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4952                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4953                         num_stripes = map->num_stripes;
4954                 } else if (mirror_num) {
4955                         stripe_index = mirror_num - 1;
4956                 } else {
4957                         mirror_num = 1;
4958                 }
4959
4960         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4961                 int factor = map->num_stripes / map->sub_stripes;
4962
4963                 stripe_index = do_div(stripe_nr, factor);
4964                 stripe_index *= map->sub_stripes;
4965
4966                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4967                         num_stripes = map->sub_stripes;
4968                 else if (rw & REQ_DISCARD)
4969                         num_stripes = min_t(u64, map->sub_stripes *
4970                                             (stripe_nr_end - stripe_nr_orig),
4971                                             map->num_stripes);
4972                 else if (mirror_num)
4973                         stripe_index += mirror_num - 1;
4974                 else {
4975                         int old_stripe_index = stripe_index;
4976                         stripe_index = find_live_mirror(fs_info, map,
4977                                               stripe_index,
4978                                               map->sub_stripes, stripe_index +
4979                                               current->pid % map->sub_stripes,
4980                                               dev_replace_is_ongoing);
4981                         mirror_num = stripe_index - old_stripe_index + 1;
4982                 }
4983
4984         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4985                                 BTRFS_BLOCK_GROUP_RAID6)) {
4986                 u64 tmp;
4987
4988                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4989                     && raid_map_ret) {
4990                         int i, rot;
4991
4992                         /* push stripe_nr back to the start of the full stripe */
4993                         stripe_nr = raid56_full_stripe_start;
4994                         do_div(stripe_nr, stripe_len);
4995
4996                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4997
4998                         /* RAID[56] write or recovery. Return all stripes */
4999                         num_stripes = map->num_stripes;
5000                         max_errors = nr_parity_stripes(map);
5001
5002                         raid_map = kmalloc_array(num_stripes, sizeof(u64),
5003                                            GFP_NOFS);
5004                         if (!raid_map) {
5005                                 ret = -ENOMEM;
5006                                 goto out;
5007                         }
5008
5009                         /* Work out the disk rotation on this stripe-set */
5010                         tmp = stripe_nr;
5011                         rot = do_div(tmp, num_stripes);
5012
5013                         /* Fill in the logical address of each stripe */
5014                         tmp = stripe_nr * nr_data_stripes(map);
5015                         for (i = 0; i < nr_data_stripes(map); i++)
5016                                 raid_map[(i+rot) % num_stripes] =
5017                                         em->start + (tmp + i) * map->stripe_len;
5018
5019                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5020                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5021                                 raid_map[(i+rot+1) % num_stripes] =
5022                                         RAID6_Q_STRIPE;
5023
5024                         *length = map->stripe_len;
5025                         stripe_index = 0;
5026                         stripe_offset = 0;
5027                 } else {
5028                         /*
5029                          * Mirror #0 or #1 means the original data block.
5030                          * Mirror #2 is RAID5 parity block.
5031                          * Mirror #3 is RAID6 Q block.
5032                          */
5033                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5034                         if (mirror_num > 1)
5035                                 stripe_index = nr_data_stripes(map) +
5036                                                 mirror_num - 2;
5037
5038                         /* We distribute the parity blocks across stripes */
5039                         tmp = stripe_nr + stripe_index;
5040                         stripe_index = do_div(tmp, map->num_stripes);
5041                 }
5042         } else {
5043                 /*
5044                  * after this do_div call, stripe_nr is the number of stripes
5045                  * on this device we have to walk to find the data, and
5046                  * stripe_index is the number of our device in the stripe array
5047                  */
5048                 stripe_index = do_div(stripe_nr, map->num_stripes);
5049                 mirror_num = stripe_index + 1;
5050         }
5051         BUG_ON(stripe_index >= map->num_stripes);
5052
5053         num_alloc_stripes = num_stripes;
5054         if (dev_replace_is_ongoing) {
5055                 if (rw & (REQ_WRITE | REQ_DISCARD))
5056                         num_alloc_stripes <<= 1;
5057                 if (rw & REQ_GET_READ_MIRRORS)
5058                         num_alloc_stripes++;
5059         }
5060         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
5061         if (!bbio) {
5062                 kfree(raid_map);
5063                 ret = -ENOMEM;
5064                 goto out;
5065         }
5066         atomic_set(&bbio->error, 0);
5067
5068         if (rw & REQ_DISCARD) {
5069                 int factor = 0;
5070                 int sub_stripes = 0;
5071                 u64 stripes_per_dev = 0;
5072                 u32 remaining_stripes = 0;
5073                 u32 last_stripe = 0;
5074
5075                 if (map->type &
5076                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5077                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5078                                 sub_stripes = 1;
5079                         else
5080                                 sub_stripes = map->sub_stripes;
5081
5082                         factor = map->num_stripes / sub_stripes;
5083                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5084                                                       stripe_nr_orig,
5085                                                       factor,
5086                                                       &remaining_stripes);
5087                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5088                         last_stripe *= sub_stripes;
5089                 }
5090
5091                 for (i = 0; i < num_stripes; i++) {
5092                         bbio->stripes[i].physical =
5093                                 map->stripes[stripe_index].physical +
5094                                 stripe_offset + stripe_nr * map->stripe_len;
5095                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5096
5097                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5098                                          BTRFS_BLOCK_GROUP_RAID10)) {
5099                                 bbio->stripes[i].length = stripes_per_dev *
5100                                                           map->stripe_len;
5101
5102                                 if (i / sub_stripes < remaining_stripes)
5103                                         bbio->stripes[i].length +=
5104                                                 map->stripe_len;
5105
5106                                 /*
5107                                  * Special for the first stripe and
5108                                  * the last stripe:
5109                                  *
5110                                  * |-------|...|-------|
5111                                  *     |----------|
5112                                  *    off     end_off
5113                                  */
5114                                 if (i < sub_stripes)
5115                                         bbio->stripes[i].length -=
5116                                                 stripe_offset;
5117
5118                                 if (stripe_index >= last_stripe &&
5119                                     stripe_index <= (last_stripe +
5120                                                      sub_stripes - 1))
5121                                         bbio->stripes[i].length -=
5122                                                 stripe_end_offset;
5123
5124                                 if (i == sub_stripes - 1)
5125                                         stripe_offset = 0;
5126                         } else
5127                                 bbio->stripes[i].length = *length;
5128
5129                         stripe_index++;
5130                         if (stripe_index == map->num_stripes) {
5131                                 /* This could only happen for RAID0/10 */
5132                                 stripe_index = 0;
5133                                 stripe_nr++;
5134                         }
5135                 }
5136         } else {
5137                 for (i = 0; i < num_stripes; i++) {
5138                         bbio->stripes[i].physical =
5139                                 map->stripes[stripe_index].physical +
5140                                 stripe_offset +
5141                                 stripe_nr * map->stripe_len;
5142                         bbio->stripes[i].dev =
5143                                 map->stripes[stripe_index].dev;
5144                         stripe_index++;
5145                 }
5146         }
5147
5148         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5149                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5150                                  BTRFS_BLOCK_GROUP_RAID10 |
5151                                  BTRFS_BLOCK_GROUP_RAID5 |
5152                                  BTRFS_BLOCK_GROUP_DUP)) {
5153                         max_errors = 1;
5154                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5155                         max_errors = 2;
5156                 }
5157         }
5158
5159         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5160             dev_replace->tgtdev != NULL) {
5161                 int index_where_to_add;
5162                 u64 srcdev_devid = dev_replace->srcdev->devid;
5163
5164                 /*
5165                  * duplicate the write operations while the dev replace
5166                  * procedure is running. Since the copying of the old disk
5167                  * to the new disk takes place at run time while the
5168                  * filesystem is mounted writable, the regular write
5169                  * operations to the old disk have to be duplicated to go
5170                  * to the new disk as well.
5171                  * Note that device->missing is handled by the caller, and
5172                  * that the write to the old disk is already set up in the
5173                  * stripes array.
5174                  */
5175                 index_where_to_add = num_stripes;
5176                 for (i = 0; i < num_stripes; i++) {
5177                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5178                                 /* write to new disk, too */
5179                                 struct btrfs_bio_stripe *new =
5180                                         bbio->stripes + index_where_to_add;
5181                                 struct btrfs_bio_stripe *old =
5182                                         bbio->stripes + i;
5183
5184                                 new->physical = old->physical;
5185                                 new->length = old->length;
5186                                 new->dev = dev_replace->tgtdev;
5187                                 index_where_to_add++;
5188                                 max_errors++;
5189                         }
5190                 }
5191                 num_stripes = index_where_to_add;
5192         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5193                    dev_replace->tgtdev != NULL) {
5194                 u64 srcdev_devid = dev_replace->srcdev->devid;
5195                 int index_srcdev = 0;
5196                 int found = 0;
5197                 u64 physical_of_found = 0;
5198
5199                 /*
5200                  * During the dev-replace procedure, the target drive can
5201                  * also be used to read data in case it is needed to repair
5202                  * a corrupt block elsewhere. This is possible if the
5203                  * requested area is left of the left cursor. In this area,
5204                  * the target drive is a full copy of the source drive.
5205                  */
5206                 for (i = 0; i < num_stripes; i++) {
5207                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5208                                 /*
5209                                  * In case of DUP, in order to keep it
5210                                  * simple, only add the mirror with the
5211                                  * lowest physical address
5212                                  */
5213                                 if (found &&
5214                                     physical_of_found <=
5215                                      bbio->stripes[i].physical)
5216                                         continue;
5217                                 index_srcdev = i;
5218                                 found = 1;
5219                                 physical_of_found = bbio->stripes[i].physical;
5220                         }
5221                 }
5222                 if (found) {
5223                         u64 length = map->stripe_len;
5224
5225                         if (physical_of_found + length <=
5226                             dev_replace->cursor_left) {
5227                                 struct btrfs_bio_stripe *tgtdev_stripe =
5228                                         bbio->stripes + num_stripes;
5229
5230                                 tgtdev_stripe->physical = physical_of_found;
5231                                 tgtdev_stripe->length =
5232                                         bbio->stripes[index_srcdev].length;
5233                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5234
5235                                 num_stripes++;
5236                         }
5237                 }
5238         }
5239
5240         *bbio_ret = bbio;
5241         bbio->num_stripes = num_stripes;
5242         bbio->max_errors = max_errors;
5243         bbio->mirror_num = mirror_num;
5244
5245         /*
5246          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5247          * mirror_num == num_stripes + 1 && dev_replace target drive is
5248          * available as a mirror
5249          */
5250         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5251                 WARN_ON(num_stripes > 1);
5252                 bbio->stripes[0].dev = dev_replace->tgtdev;
5253                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5254                 bbio->mirror_num = map->num_stripes + 1;
5255         }
5256         if (raid_map) {
5257                 sort_parity_stripes(bbio, raid_map);
5258                 *raid_map_ret = raid_map;
5259         }
5260 out:
5261         if (dev_replace_is_ongoing)
5262                 btrfs_dev_replace_unlock(dev_replace);
5263         free_extent_map(em);
5264         return ret;
5265 }
5266
5267 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5268                       u64 logical, u64 *length,
5269                       struct btrfs_bio **bbio_ret, int mirror_num)
5270 {
5271         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5272                                  mirror_num, NULL);
5273 }
5274
5275 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5276                      u64 chunk_start, u64 physical, u64 devid,
5277                      u64 **logical, int *naddrs, int *stripe_len)
5278 {
5279         struct extent_map_tree *em_tree = &map_tree->map_tree;
5280         struct extent_map *em;
5281         struct map_lookup *map;
5282         u64 *buf;
5283         u64 bytenr;
5284         u64 length;
5285         u64 stripe_nr;
5286         u64 rmap_len;
5287         int i, j, nr = 0;
5288
5289         read_lock(&em_tree->lock);
5290         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5291         read_unlock(&em_tree->lock);
5292
5293         if (!em) {
5294                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5295                        chunk_start);
5296                 return -EIO;
5297         }
5298
5299         if (em->start != chunk_start) {
5300                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5301                        em->start, chunk_start);
5302                 free_extent_map(em);
5303                 return -EIO;
5304         }
5305         map = (struct map_lookup *)em->bdev;
5306
5307         length = em->len;
5308         rmap_len = map->stripe_len;
5309
5310         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5311                 do_div(length, map->num_stripes / map->sub_stripes);
5312         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5313                 do_div(length, map->num_stripes);
5314         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5315                               BTRFS_BLOCK_GROUP_RAID6)) {
5316                 do_div(length, nr_data_stripes(map));
5317                 rmap_len = map->stripe_len * nr_data_stripes(map);
5318         }
5319
5320         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5321         BUG_ON(!buf); /* -ENOMEM */
5322
5323         for (i = 0; i < map->num_stripes; i++) {
5324                 if (devid && map->stripes[i].dev->devid != devid)
5325                         continue;
5326                 if (map->stripes[i].physical > physical ||
5327                     map->stripes[i].physical + length <= physical)
5328                         continue;
5329
5330                 stripe_nr = physical - map->stripes[i].physical;
5331                 do_div(stripe_nr, map->stripe_len);
5332
5333                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5334                         stripe_nr = stripe_nr * map->num_stripes + i;
5335                         do_div(stripe_nr, map->sub_stripes);
5336                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5337                         stripe_nr = stripe_nr * map->num_stripes + i;
5338                 } /* else if RAID[56], multiply by nr_data_stripes().
5339                    * Alternatively, just use rmap_len below instead of
5340                    * map->stripe_len */
5341
5342                 bytenr = chunk_start + stripe_nr * rmap_len;
5343                 WARN_ON(nr >= map->num_stripes);
5344                 for (j = 0; j < nr; j++) {
5345                         if (buf[j] == bytenr)
5346                                 break;
5347                 }
5348                 if (j == nr) {
5349                         WARN_ON(nr >= map->num_stripes);
5350                         buf[nr++] = bytenr;
5351                 }
5352         }
5353
5354         *logical = buf;
5355         *naddrs = nr;
5356         *stripe_len = rmap_len;
5357
5358         free_extent_map(em);
5359         return 0;
5360 }
5361
5362 static void btrfs_end_bio(struct bio *bio, int err)
5363 {
5364         struct btrfs_bio *bbio = bio->bi_private;
5365         struct btrfs_device *dev = bbio->stripes[0].dev;
5366         int is_orig_bio = 0;
5367
5368         if (err) {
5369                 atomic_inc(&bbio->error);
5370                 if (err == -EIO || err == -EREMOTEIO) {
5371                         unsigned int stripe_index =
5372                                 btrfs_io_bio(bio)->stripe_index;
5373
5374                         BUG_ON(stripe_index >= bbio->num_stripes);
5375                         dev = bbio->stripes[stripe_index].dev;
5376                         if (dev->bdev) {
5377                                 if (bio->bi_rw & WRITE)
5378                                         btrfs_dev_stat_inc(dev,
5379                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5380                                 else
5381                                         btrfs_dev_stat_inc(dev,
5382                                                 BTRFS_DEV_STAT_READ_ERRS);
5383                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5384                                         btrfs_dev_stat_inc(dev,
5385                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5386                                 btrfs_dev_stat_print_on_error(dev);
5387                         }
5388                 }
5389         }
5390
5391         if (bio == bbio->orig_bio)
5392                 is_orig_bio = 1;
5393
5394         btrfs_bio_counter_dec(bbio->fs_info);
5395
5396         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5397                 if (!is_orig_bio) {
5398                         bio_put(bio);
5399                         bio = bbio->orig_bio;
5400                 }
5401
5402                 /*
5403                  * We have original bio now. So increment bi_remaining to
5404                  * account for it in endio
5405                  */
5406                 atomic_inc(&bio->bi_remaining);
5407
5408                 bio->bi_private = bbio->private;
5409                 bio->bi_end_io = bbio->end_io;
5410                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5411                 /* only send an error to the higher layers if it is
5412                  * beyond the tolerance of the btrfs bio
5413                  */
5414                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5415                         err = -EIO;
5416                 } else {
5417                         /*
5418                          * this bio is actually up to date, we didn't
5419                          * go over the max number of errors
5420                          */
5421                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5422                         err = 0;
5423                 }
5424                 kfree(bbio);
5425
5426                 bio_endio(bio, err);
5427         } else if (!is_orig_bio) {
5428                 bio_put(bio);
5429         }
5430 }
5431
5432 /*
5433  * see run_scheduled_bios for a description of why bios are collected for
5434  * async submit.
5435  *
5436  * This will add one bio to the pending list for a device and make sure
5437  * the work struct is scheduled.
5438  */
5439 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5440                                         struct btrfs_device *device,
5441                                         int rw, struct bio *bio)
5442 {
5443         int should_queue = 1;
5444         struct btrfs_pending_bios *pending_bios;
5445
5446         if (device->missing || !device->bdev) {
5447                 bio_endio(bio, -EIO);
5448                 return;
5449         }
5450
5451         /* don't bother with additional async steps for reads, right now */
5452         if (!(rw & REQ_WRITE)) {
5453                 bio_get(bio);
5454                 btrfsic_submit_bio(rw, bio);
5455                 bio_put(bio);
5456                 return;
5457         }
5458
5459         /*
5460          * nr_async_bios allows us to reliably return congestion to the
5461          * higher layers.  Otherwise, the async bio makes it appear we have
5462          * made progress against dirty pages when we've really just put it
5463          * on a queue for later
5464          */
5465         atomic_inc(&root->fs_info->nr_async_bios);
5466         WARN_ON(bio->bi_next);
5467         bio->bi_next = NULL;
5468         bio->bi_rw |= rw;
5469
5470         spin_lock(&device->io_lock);
5471         if (bio->bi_rw & REQ_SYNC)
5472                 pending_bios = &device->pending_sync_bios;
5473         else
5474                 pending_bios = &device->pending_bios;
5475
5476         if (pending_bios->tail)
5477                 pending_bios->tail->bi_next = bio;
5478
5479         pending_bios->tail = bio;
5480         if (!pending_bios->head)
5481                 pending_bios->head = bio;
5482         if (device->running_pending)
5483                 should_queue = 0;
5484
5485         spin_unlock(&device->io_lock);
5486
5487         if (should_queue)
5488                 btrfs_queue_work(root->fs_info->submit_workers,
5489                                  &device->work);
5490 }
5491
5492 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5493                        sector_t sector)
5494 {
5495         struct bio_vec *prev;
5496         struct request_queue *q = bdev_get_queue(bdev);
5497         unsigned int max_sectors = queue_max_sectors(q);
5498         struct bvec_merge_data bvm = {
5499                 .bi_bdev = bdev,
5500                 .bi_sector = sector,
5501                 .bi_rw = bio->bi_rw,
5502         };
5503
5504         if (WARN_ON(bio->bi_vcnt == 0))
5505                 return 1;
5506
5507         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5508         if (bio_sectors(bio) > max_sectors)
5509                 return 0;
5510
5511         if (!q->merge_bvec_fn)
5512                 return 1;
5513
5514         bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5515         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5516                 return 0;
5517         return 1;
5518 }
5519
5520 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5521                               struct bio *bio, u64 physical, int dev_nr,
5522                               int rw, int async)
5523 {
5524         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5525
5526         bio->bi_private = bbio;
5527         btrfs_io_bio(bio)->stripe_index = dev_nr;
5528         bio->bi_end_io = btrfs_end_bio;
5529         bio->bi_iter.bi_sector = physical >> 9;
5530 #ifdef DEBUG
5531         {
5532                 struct rcu_string *name;
5533
5534                 rcu_read_lock();
5535                 name = rcu_dereference(dev->name);
5536                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5537                          "(%s id %llu), size=%u\n", rw,
5538                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5539                          name->str, dev->devid, bio->bi_size);
5540                 rcu_read_unlock();
5541         }
5542 #endif
5543         bio->bi_bdev = dev->bdev;
5544
5545         btrfs_bio_counter_inc_noblocked(root->fs_info);
5546
5547         if (async)
5548                 btrfs_schedule_bio(root, dev, rw, bio);
5549         else
5550                 btrfsic_submit_bio(rw, bio);
5551 }
5552
5553 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5554                               struct bio *first_bio, struct btrfs_device *dev,
5555                               int dev_nr, int rw, int async)
5556 {
5557         struct bio_vec *bvec = first_bio->bi_io_vec;
5558         struct bio *bio;
5559         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5560         u64 physical = bbio->stripes[dev_nr].physical;
5561
5562 again:
5563         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5564         if (!bio)
5565                 return -ENOMEM;
5566
5567         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5568                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5569                                  bvec->bv_offset) < bvec->bv_len) {
5570                         u64 len = bio->bi_iter.bi_size;
5571
5572                         atomic_inc(&bbio->stripes_pending);
5573                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5574                                           rw, async);
5575                         physical += len;
5576                         goto again;
5577                 }
5578                 bvec++;
5579         }
5580
5581         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5582         return 0;
5583 }
5584
5585 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5586 {
5587         atomic_inc(&bbio->error);
5588         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5589                 bio->bi_private = bbio->private;
5590                 bio->bi_end_io = bbio->end_io;
5591                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5592                 bio->bi_iter.bi_sector = logical >> 9;
5593                 kfree(bbio);
5594                 bio_endio(bio, -EIO);
5595         }
5596 }
5597
5598 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5599                   int mirror_num, int async_submit)
5600 {
5601         struct btrfs_device *dev;
5602         struct bio *first_bio = bio;
5603         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5604         u64 length = 0;
5605         u64 map_length;
5606         u64 *raid_map = NULL;
5607         int ret;
5608         int dev_nr = 0;
5609         int total_devs = 1;
5610         struct btrfs_bio *bbio = NULL;
5611
5612         length = bio->bi_iter.bi_size;
5613         map_length = length;
5614
5615         btrfs_bio_counter_inc_blocked(root->fs_info);
5616         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5617                               mirror_num, &raid_map);
5618         if (ret) {
5619                 btrfs_bio_counter_dec(root->fs_info);
5620                 return ret;
5621         }
5622
5623         total_devs = bbio->num_stripes;
5624         bbio->orig_bio = first_bio;
5625         bbio->private = first_bio->bi_private;
5626         bbio->end_io = first_bio->bi_end_io;
5627         bbio->fs_info = root->fs_info;
5628         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5629
5630         if (raid_map) {
5631                 /* In this case, map_length has been set to the length of
5632                    a single stripe; not the whole write */
5633                 if (rw & WRITE) {
5634                         ret = raid56_parity_write(root, bio, bbio,
5635                                                   raid_map, map_length);
5636                 } else {
5637                         ret = raid56_parity_recover(root, bio, bbio,
5638                                                     raid_map, map_length,
5639                                                     mirror_num);
5640                 }
5641                 /*
5642                  * FIXME, replace dosen't support raid56 yet, please fix
5643                  * it in the future.
5644                  */
5645                 btrfs_bio_counter_dec(root->fs_info);
5646                 return ret;
5647         }
5648
5649         if (map_length < length) {
5650                 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5651                         logical, length, map_length);
5652                 BUG();
5653         }
5654
5655         while (dev_nr < total_devs) {
5656                 dev = bbio->stripes[dev_nr].dev;
5657                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5658                         bbio_error(bbio, first_bio, logical);
5659                         dev_nr++;
5660                         continue;
5661                 }
5662
5663                 /*
5664                  * Check and see if we're ok with this bio based on it's size
5665                  * and offset with the given device.
5666                  */
5667                 if (!bio_size_ok(dev->bdev, first_bio,
5668                                  bbio->stripes[dev_nr].physical >> 9)) {
5669                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5670                                                  dev_nr, rw, async_submit);
5671                         BUG_ON(ret);
5672                         dev_nr++;
5673                         continue;
5674                 }
5675
5676                 if (dev_nr < total_devs - 1) {
5677                         bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5678                         BUG_ON(!bio); /* -ENOMEM */
5679                 } else {
5680                         bio = first_bio;
5681                 }
5682
5683                 submit_stripe_bio(root, bbio, bio,
5684                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5685                                   async_submit);
5686                 dev_nr++;
5687         }
5688         btrfs_bio_counter_dec(root->fs_info);
5689         return 0;
5690 }
5691
5692 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5693                                        u8 *uuid, u8 *fsid)
5694 {
5695         struct btrfs_device *device;
5696         struct btrfs_fs_devices *cur_devices;
5697
5698         cur_devices = fs_info->fs_devices;
5699         while (cur_devices) {
5700                 if (!fsid ||
5701                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5702                         device = __find_device(&cur_devices->devices,
5703                                                devid, uuid);
5704                         if (device)
5705                                 return device;
5706                 }
5707                 cur_devices = cur_devices->seed;
5708         }
5709         return NULL;
5710 }
5711
5712 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5713                                             u64 devid, u8 *dev_uuid)
5714 {
5715         struct btrfs_device *device;
5716         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5717
5718         device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5719         if (IS_ERR(device))
5720                 return NULL;
5721
5722         list_add(&device->dev_list, &fs_devices->devices);
5723         device->fs_devices = fs_devices;
5724         fs_devices->num_devices++;
5725
5726         device->missing = 1;
5727         fs_devices->missing_devices++;
5728
5729         return device;
5730 }
5731
5732 /**
5733  * btrfs_alloc_device - allocate struct btrfs_device
5734  * @fs_info:    used only for generating a new devid, can be NULL if
5735  *              devid is provided (i.e. @devid != NULL).
5736  * @devid:      a pointer to devid for this device.  If NULL a new devid
5737  *              is generated.
5738  * @uuid:       a pointer to UUID for this device.  If NULL a new UUID
5739  *              is generated.
5740  *
5741  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5742  * on error.  Returned struct is not linked onto any lists and can be
5743  * destroyed with kfree() right away.
5744  */
5745 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5746                                         const u64 *devid,
5747                                         const u8 *uuid)
5748 {
5749         struct btrfs_device *dev;
5750         u64 tmp;
5751
5752         if (WARN_ON(!devid && !fs_info))
5753                 return ERR_PTR(-EINVAL);
5754
5755         dev = __alloc_device();
5756         if (IS_ERR(dev))
5757                 return dev;
5758
5759         if (devid)
5760                 tmp = *devid;
5761         else {
5762                 int ret;
5763
5764                 ret = find_next_devid(fs_info, &tmp);
5765                 if (ret) {
5766                         kfree(dev);
5767                         return ERR_PTR(ret);
5768                 }
5769         }
5770         dev->devid = tmp;
5771
5772         if (uuid)
5773                 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5774         else
5775                 generate_random_uuid(dev->uuid);
5776
5777         btrfs_init_work(&dev->work, pending_bios_fn, NULL, NULL);
5778
5779         return dev;
5780 }
5781
5782 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5783                           struct extent_buffer *leaf,
5784                           struct btrfs_chunk *chunk)
5785 {
5786         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5787         struct map_lookup *map;
5788         struct extent_map *em;
5789         u64 logical;
5790         u64 length;
5791         u64 devid;
5792         u8 uuid[BTRFS_UUID_SIZE];
5793         int num_stripes;
5794         int ret;
5795         int i;
5796
5797         logical = key->offset;
5798         length = btrfs_chunk_length(leaf, chunk);
5799
5800         read_lock(&map_tree->map_tree.lock);
5801         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5802         read_unlock(&map_tree->map_tree.lock);
5803
5804         /* already mapped? */
5805         if (em && em->start <= logical && em->start + em->len > logical) {
5806                 free_extent_map(em);
5807                 return 0;
5808         } else if (em) {
5809                 free_extent_map(em);
5810         }
5811
5812         em = alloc_extent_map();
5813         if (!em)
5814                 return -ENOMEM;
5815         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5816         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5817         if (!map) {
5818                 free_extent_map(em);
5819                 return -ENOMEM;
5820         }
5821
5822         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5823         em->bdev = (struct block_device *)map;
5824         em->start = logical;
5825         em->len = length;
5826         em->orig_start = 0;
5827         em->block_start = 0;
5828         em->block_len = em->len;
5829
5830         map->num_stripes = num_stripes;
5831         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5832         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5833         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5834         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5835         map->type = btrfs_chunk_type(leaf, chunk);
5836         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5837         for (i = 0; i < num_stripes; i++) {
5838                 map->stripes[i].physical =
5839                         btrfs_stripe_offset_nr(leaf, chunk, i);
5840                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5841                 read_extent_buffer(leaf, uuid, (unsigned long)
5842                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5843                                    BTRFS_UUID_SIZE);
5844                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5845                                                         uuid, NULL);
5846                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5847                         free_extent_map(em);
5848                         return -EIO;
5849                 }
5850                 if (!map->stripes[i].dev) {
5851                         map->stripes[i].dev =
5852                                 add_missing_dev(root, devid, uuid);
5853                         if (!map->stripes[i].dev) {
5854                                 free_extent_map(em);
5855                                 return -EIO;
5856                         }
5857                 }
5858                 map->stripes[i].dev->in_fs_metadata = 1;
5859         }
5860
5861         write_lock(&map_tree->map_tree.lock);
5862         ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5863         write_unlock(&map_tree->map_tree.lock);
5864         BUG_ON(ret); /* Tree corruption */
5865         free_extent_map(em);
5866
5867         return 0;
5868 }
5869
5870 static void fill_device_from_item(struct extent_buffer *leaf,
5871                                  struct btrfs_dev_item *dev_item,
5872                                  struct btrfs_device *device)
5873 {
5874         unsigned long ptr;
5875
5876         device->devid = btrfs_device_id(leaf, dev_item);
5877         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5878         device->total_bytes = device->disk_total_bytes;
5879         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5880         device->type = btrfs_device_type(leaf, dev_item);
5881         device->io_align = btrfs_device_io_align(leaf, dev_item);
5882         device->io_width = btrfs_device_io_width(leaf, dev_item);
5883         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5884         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5885         device->is_tgtdev_for_dev_replace = 0;
5886
5887         ptr = btrfs_device_uuid(dev_item);
5888         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5889 }
5890
5891 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5892 {
5893         struct btrfs_fs_devices *fs_devices;
5894         int ret;
5895
5896         BUG_ON(!mutex_is_locked(&uuid_mutex));
5897
5898         fs_devices = root->fs_info->fs_devices->seed;
5899         while (fs_devices) {
5900                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5901                         ret = 0;
5902                         goto out;
5903                 }
5904                 fs_devices = fs_devices->seed;
5905         }
5906
5907         fs_devices = find_fsid(fsid);
5908         if (!fs_devices) {
5909                 ret = -ENOENT;
5910                 goto out;
5911         }
5912
5913         fs_devices = clone_fs_devices(fs_devices);
5914         if (IS_ERR(fs_devices)) {
5915                 ret = PTR_ERR(fs_devices);
5916                 goto out;
5917         }
5918
5919         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5920                                    root->fs_info->bdev_holder);
5921         if (ret) {
5922                 free_fs_devices(fs_devices);
5923                 goto out;
5924         }
5925
5926         if (!fs_devices->seeding) {
5927                 __btrfs_close_devices(fs_devices);
5928                 free_fs_devices(fs_devices);
5929                 ret = -EINVAL;
5930                 goto out;
5931         }
5932
5933         fs_devices->seed = root->fs_info->fs_devices->seed;
5934         root->fs_info->fs_devices->seed = fs_devices;
5935 out:
5936         return ret;
5937 }
5938
5939 static int read_one_dev(struct btrfs_root *root,
5940                         struct extent_buffer *leaf,
5941                         struct btrfs_dev_item *dev_item)
5942 {
5943         struct btrfs_device *device;
5944         u64 devid;
5945         int ret;
5946         u8 fs_uuid[BTRFS_UUID_SIZE];
5947         u8 dev_uuid[BTRFS_UUID_SIZE];
5948
5949         devid = btrfs_device_id(leaf, dev_item);
5950         read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5951                            BTRFS_UUID_SIZE);
5952         read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5953                            BTRFS_UUID_SIZE);
5954
5955         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5956                 ret = open_seed_devices(root, fs_uuid);
5957                 if (ret && !btrfs_test_opt(root, DEGRADED))
5958                         return ret;
5959         }
5960
5961         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5962         if (!device || !device->bdev) {
5963                 if (!btrfs_test_opt(root, DEGRADED))
5964                         return -EIO;
5965
5966                 if (!device) {
5967                         btrfs_warn(root->fs_info, "devid %llu missing", devid);
5968                         device = add_missing_dev(root, devid, dev_uuid);
5969                         if (!device)
5970                                 return -ENOMEM;
5971                 } else if (!device->missing) {
5972                         /*
5973                          * this happens when a device that was properly setup
5974                          * in the device info lists suddenly goes bad.
5975                          * device->bdev is NULL, and so we have to set
5976                          * device->missing to one here
5977                          */
5978                         root->fs_info->fs_devices->missing_devices++;
5979                         device->missing = 1;
5980                 }
5981         }
5982
5983         if (device->fs_devices != root->fs_info->fs_devices) {
5984                 BUG_ON(device->writeable);
5985                 if (device->generation !=
5986                     btrfs_device_generation(leaf, dev_item))
5987                         return -EINVAL;
5988         }
5989
5990         fill_device_from_item(leaf, dev_item, device);
5991         device->in_fs_metadata = 1;
5992         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5993                 device->fs_devices->total_rw_bytes += device->total_bytes;
5994                 spin_lock(&root->fs_info->free_chunk_lock);
5995                 root->fs_info->free_chunk_space += device->total_bytes -
5996                         device->bytes_used;
5997                 spin_unlock(&root->fs_info->free_chunk_lock);
5998         }
5999         ret = 0;
6000         return ret;
6001 }
6002
6003 int btrfs_read_sys_array(struct btrfs_root *root)
6004 {
6005         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6006         struct extent_buffer *sb;
6007         struct btrfs_disk_key *disk_key;
6008         struct btrfs_chunk *chunk;
6009         u8 *ptr;
6010         unsigned long sb_ptr;
6011         int ret = 0;
6012         u32 num_stripes;
6013         u32 array_size;
6014         u32 len = 0;
6015         u32 cur;
6016         struct btrfs_key key;
6017
6018         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
6019                                           BTRFS_SUPER_INFO_SIZE);
6020         if (!sb)
6021                 return -ENOMEM;
6022         btrfs_set_buffer_uptodate(sb);
6023         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6024         /*
6025          * The sb extent buffer is artifical and just used to read the system array.
6026          * btrfs_set_buffer_uptodate() call does not properly mark all it's
6027          * pages up-to-date when the page is larger: extent does not cover the
6028          * whole page and consequently check_page_uptodate does not find all
6029          * the page's extents up-to-date (the hole beyond sb),
6030          * write_extent_buffer then triggers a WARN_ON.
6031          *
6032          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6033          * but sb spans only this function. Add an explicit SetPageUptodate call
6034          * to silence the warning eg. on PowerPC 64.
6035          */
6036         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6037                 SetPageUptodate(sb->pages[0]);
6038
6039         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6040         array_size = btrfs_super_sys_array_size(super_copy);
6041
6042         ptr = super_copy->sys_chunk_array;
6043         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
6044         cur = 0;
6045
6046         while (cur < array_size) {
6047                 disk_key = (struct btrfs_disk_key *)ptr;
6048                 btrfs_disk_key_to_cpu(&key, disk_key);
6049
6050                 len = sizeof(*disk_key); ptr += len;
6051                 sb_ptr += len;
6052                 cur += len;
6053
6054                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6055                         chunk = (struct btrfs_chunk *)sb_ptr;
6056                         ret = read_one_chunk(root, &key, sb, chunk);
6057                         if (ret)
6058                                 break;
6059                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6060                         len = btrfs_chunk_item_size(num_stripes);
6061                 } else {
6062                         ret = -EIO;
6063                         break;
6064                 }
6065                 ptr += len;
6066                 sb_ptr += len;
6067                 cur += len;
6068         }
6069         free_extent_buffer(sb);
6070         return ret;
6071 }
6072
6073 int btrfs_read_chunk_tree(struct btrfs_root *root)
6074 {
6075         struct btrfs_path *path;
6076         struct extent_buffer *leaf;
6077         struct btrfs_key key;
6078         struct btrfs_key found_key;
6079         int ret;
6080         int slot;
6081
6082         root = root->fs_info->chunk_root;
6083
6084         path = btrfs_alloc_path();
6085         if (!path)
6086                 return -ENOMEM;
6087
6088         mutex_lock(&uuid_mutex);
6089         lock_chunks(root);
6090
6091         /*
6092          * Read all device items, and then all the chunk items. All
6093          * device items are found before any chunk item (their object id
6094          * is smaller than the lowest possible object id for a chunk
6095          * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6096          */
6097         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6098         key.offset = 0;
6099         key.type = 0;
6100         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6101         if (ret < 0)
6102                 goto error;
6103         while (1) {
6104                 leaf = path->nodes[0];
6105                 slot = path->slots[0];
6106                 if (slot >= btrfs_header_nritems(leaf)) {
6107                         ret = btrfs_next_leaf(root, path);
6108                         if (ret == 0)
6109                                 continue;
6110                         if (ret < 0)
6111                                 goto error;
6112                         break;
6113                 }
6114                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6115                 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6116                         struct btrfs_dev_item *dev_item;
6117                         dev_item = btrfs_item_ptr(leaf, slot,
6118                                                   struct btrfs_dev_item);
6119                         ret = read_one_dev(root, leaf, dev_item);
6120                         if (ret)
6121                                 goto error;
6122                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6123                         struct btrfs_chunk *chunk;
6124                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6125                         ret = read_one_chunk(root, &found_key, leaf, chunk);
6126                         if (ret)
6127                                 goto error;
6128                 }
6129                 path->slots[0]++;
6130         }
6131         ret = 0;
6132 error:
6133         unlock_chunks(root);
6134         mutex_unlock(&uuid_mutex);
6135
6136         btrfs_free_path(path);
6137         return ret;
6138 }
6139
6140 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6141 {
6142         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6143         struct btrfs_device *device;
6144
6145         while (fs_devices) {
6146                 mutex_lock(&fs_devices->device_list_mutex);
6147                 list_for_each_entry(device, &fs_devices->devices, dev_list)
6148                         device->dev_root = fs_info->dev_root;
6149                 mutex_unlock(&fs_devices->device_list_mutex);
6150
6151                 fs_devices = fs_devices->seed;
6152         }
6153 }
6154
6155 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6156 {
6157         int i;
6158
6159         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6160                 btrfs_dev_stat_reset(dev, i);
6161 }
6162
6163 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6164 {
6165         struct btrfs_key key;
6166         struct btrfs_key found_key;
6167         struct btrfs_root *dev_root = fs_info->dev_root;
6168         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6169         struct extent_buffer *eb;
6170         int slot;
6171         int ret = 0;
6172         struct btrfs_device *device;
6173         struct btrfs_path *path = NULL;
6174         int i;
6175
6176         path = btrfs_alloc_path();
6177         if (!path) {
6178                 ret = -ENOMEM;
6179                 goto out;
6180         }
6181
6182         mutex_lock(&fs_devices->device_list_mutex);
6183         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6184                 int item_size;
6185                 struct btrfs_dev_stats_item *ptr;
6186
6187                 key.objectid = 0;
6188                 key.type = BTRFS_DEV_STATS_KEY;
6189                 key.offset = device->devid;
6190                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6191                 if (ret) {
6192                         __btrfs_reset_dev_stats(device);
6193                         device->dev_stats_valid = 1;
6194                         btrfs_release_path(path);
6195                         continue;
6196                 }
6197                 slot = path->slots[0];
6198                 eb = path->nodes[0];
6199                 btrfs_item_key_to_cpu(eb, &found_key, slot);
6200                 item_size = btrfs_item_size_nr(eb, slot);
6201
6202                 ptr = btrfs_item_ptr(eb, slot,
6203                                      struct btrfs_dev_stats_item);
6204
6205                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6206                         if (item_size >= (1 + i) * sizeof(__le64))
6207                                 btrfs_dev_stat_set(device, i,
6208                                         btrfs_dev_stats_value(eb, ptr, i));
6209                         else
6210                                 btrfs_dev_stat_reset(device, i);
6211                 }
6212
6213                 device->dev_stats_valid = 1;
6214                 btrfs_dev_stat_print_on_load(device);
6215                 btrfs_release_path(path);
6216         }
6217         mutex_unlock(&fs_devices->device_list_mutex);
6218
6219 out:
6220         btrfs_free_path(path);
6221         return ret < 0 ? ret : 0;
6222 }
6223
6224 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6225                                 struct btrfs_root *dev_root,
6226                                 struct btrfs_device *device)
6227 {
6228         struct btrfs_path *path;
6229         struct btrfs_key key;
6230         struct extent_buffer *eb;
6231         struct btrfs_dev_stats_item *ptr;
6232         int ret;
6233         int i;
6234
6235         key.objectid = 0;
6236         key.type = BTRFS_DEV_STATS_KEY;
6237         key.offset = device->devid;
6238
6239         path = btrfs_alloc_path();
6240         BUG_ON(!path);
6241         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6242         if (ret < 0) {
6243                 printk_in_rcu(KERN_WARNING "BTRFS: "
6244                         "error %d while searching for dev_stats item for device %s!\n",
6245                               ret, rcu_str_deref(device->name));
6246                 goto out;
6247         }
6248
6249         if (ret == 0 &&
6250             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6251                 /* need to delete old one and insert a new one */
6252                 ret = btrfs_del_item(trans, dev_root, path);
6253                 if (ret != 0) {
6254                         printk_in_rcu(KERN_WARNING "BTRFS: "
6255                                 "delete too small dev_stats item for device %s failed %d!\n",
6256                                       rcu_str_deref(device->name), ret);
6257                         goto out;
6258                 }
6259                 ret = 1;
6260         }
6261
6262         if (ret == 1) {
6263                 /* need to insert a new item */
6264                 btrfs_release_path(path);
6265                 ret = btrfs_insert_empty_item(trans, dev_root, path,
6266                                               &key, sizeof(*ptr));
6267                 if (ret < 0) {
6268                         printk_in_rcu(KERN_WARNING "BTRFS: "
6269                                           "insert dev_stats item for device %s failed %d!\n",
6270                                       rcu_str_deref(device->name), ret);
6271                         goto out;
6272                 }
6273         }
6274
6275         eb = path->nodes[0];
6276         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6277         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6278                 btrfs_set_dev_stats_value(eb, ptr, i,
6279                                           btrfs_dev_stat_read(device, i));
6280         btrfs_mark_buffer_dirty(eb);
6281
6282 out:
6283         btrfs_free_path(path);
6284         return ret;
6285 }
6286
6287 /*
6288  * called from commit_transaction. Writes all changed device stats to disk.
6289  */
6290 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6291                         struct btrfs_fs_info *fs_info)
6292 {
6293         struct btrfs_root *dev_root = fs_info->dev_root;
6294         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6295         struct btrfs_device *device;
6296         int ret = 0;
6297
6298         mutex_lock(&fs_devices->device_list_mutex);
6299         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6300                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
6301                         continue;
6302
6303                 ret = update_dev_stat_item(trans, dev_root, device);
6304                 if (!ret)
6305                         device->dev_stats_dirty = 0;
6306         }
6307         mutex_unlock(&fs_devices->device_list_mutex);
6308
6309         return ret;
6310 }
6311
6312 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6313 {
6314         btrfs_dev_stat_inc(dev, index);
6315         btrfs_dev_stat_print_on_error(dev);
6316 }
6317
6318 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6319 {
6320         if (!dev->dev_stats_valid)
6321                 return;
6322         printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6323                            "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6324                            rcu_str_deref(dev->name),
6325                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6326                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6327                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6328                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6329                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6330 }
6331
6332 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6333 {
6334         int i;
6335
6336         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6337                 if (btrfs_dev_stat_read(dev, i) != 0)
6338                         break;
6339         if (i == BTRFS_DEV_STAT_VALUES_MAX)
6340                 return; /* all values == 0, suppress message */
6341
6342         printk_in_rcu(KERN_INFO "BTRFS: "
6343                    "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6344                rcu_str_deref(dev->name),
6345                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6346                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6347                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6348                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6349                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6350 }
6351
6352 int btrfs_get_dev_stats(struct btrfs_root *root,
6353                         struct btrfs_ioctl_get_dev_stats *stats)
6354 {
6355         struct btrfs_device *dev;
6356         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6357         int i;
6358
6359         mutex_lock(&fs_devices->device_list_mutex);
6360         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6361         mutex_unlock(&fs_devices->device_list_mutex);
6362
6363         if (!dev) {
6364                 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6365                 return -ENODEV;
6366         } else if (!dev->dev_stats_valid) {
6367                 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6368                 return -ENODEV;
6369         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6370                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6371                         if (stats->nr_items > i)
6372                                 stats->values[i] =
6373                                         btrfs_dev_stat_read_and_reset(dev, i);
6374                         else
6375                                 btrfs_dev_stat_reset(dev, i);
6376                 }
6377         } else {
6378                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6379                         if (stats->nr_items > i)
6380                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
6381         }
6382         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6383                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6384         return 0;
6385 }
6386
6387 int btrfs_scratch_superblock(struct btrfs_device *device)
6388 {
6389         struct buffer_head *bh;
6390         struct btrfs_super_block *disk_super;
6391
6392         bh = btrfs_read_dev_super(device->bdev);
6393         if (!bh)
6394                 return -EINVAL;
6395         disk_super = (struct btrfs_super_block *)bh->b_data;
6396
6397         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6398         set_buffer_dirty(bh);
6399         sync_dirty_buffer(bh);
6400         brelse(bh);
6401
6402         return 0;
6403 }